Testing the bots for dice bias
A Modern Caveman
I'm taking JellyFish on level 7 and having it play a series of 500-point
matches against itself where I use Ctrl-A to have the game play out each time.
I would advise others who are curious to run their own simulations and see what
results they get. This is an empirical way of seeing if the dice are balanced
between the two bots (Jellyfish and Player2, who I call Jelly2).
Ray Gordon, GENIUS
http://www.cybersheet.com/library.html
Ric
about fifteen minutes. I don't see that the result tells me anything about
the dice.
Ric
"A Modern Caveman" <amodern...@aol.com> wrote in message
news:20020616184033...@mb-mo.aol.com...
A Modern Caveman
>about fifteen minutes. I don't see that the result tells me anything about
>the dice.
I'm using 500-point matches. So far I've done 35 of them.
Ric
out the cube and playing skill and luck, you don't get a clear picture of
the dice, do you? Why should points scored/matches won tell you anything
about the dice? If Jf and Jf2 in my 1007 game sample were exactly equal in
playing strength, and the dice were reversed, does that result (assuming
it's the same) prove anything about the dice?
Is your focus on the dice, or on the Fish's play? If the dice, then better
to print out 100,000 rolls of the dice from the rng and analyze that
statistically, I would think. That tells you if the dice/rng are good or
bad.
Anyway, aside from all that, what's the starting point of the exercise? Is
it that Jellyfish cheats? You didn't say that, but this kind of post usually
seems to spring from that well, and that well is dry.
What does "balanced between the bots" mean? How do you define that? Does it
mean each bot should get the same rolls? The same number of certain classes
of rolls (doubles, non-doubles, ones, twos, threes, etcetera)? If you want
to run an experiment, shouldn't you carefully define your goals,
assumptions, terms, and methodology? All you've said here is that you're
running a long series of unrealistically long matches and that is somehow
supposed to prove something ill-defined about JF's rng.
Isn't the whole deal you've postulated kind of meaningless in the short run
of dice that marks most matches (even the 25 point matches)?
"A Modern Caveman" <amodern...@aol.com> wrote in message
news:20020616200415...@mb-mo.aol.com...
A Modern Caveman
>out the cube and playing skill and luck, you don't get a clear picture of
>the dice, do you? Why should points scored/matches won tell you anything
>about the dice? If Jf and Jf2 in my 1007 game sample were exactly equal in
>playing strength, and the dice were reversed, does that result (assuming
>it's the same) prove anything about the dice?
1,007 games isn't enough for what I'm doing here.
>Is your focus on the dice, or on the Fish's play? If the dice, then better
>to print out 100,000 rolls of the dice from the rng and analyze that
>statistically, I would think. That tells you if the dice/rng are good or
>bad.
>
>Anyway, aside from all that, what's the starting point of the exercise? Is
>it that Jellyfish cheats?
No. It's that IF there is a bias in the dice, intentional or not, it will show
up in a series of 500-point matches.
A 500-point match really isn't luck, and total points won should be roughly
equal.
>You didn't say that, but this kind of post usually
>seems to spring from that well, and that well is dry.
>
>What does "balanced between the bots" mean? How do you define that? Does it
>mean each bot should get the same rolls?
No. It means that there is no reason to expect either side to win more often
than the other.
I remember when I had a commodore 64 that had a random-number generator. I
used it to make a dice simulator and after leaving it running for days, was
surprised to see the even numbers coming up much more often than the odd ones.
>The same number of certain classes
>of rolls (doubles, non-doubles, ones, twos, threes, etcetera)? If you want
>to run an experiment, shouldn't you carefully define your goals,
>assumptions, terms, and methodology?
Simple: play a series of 500-point matches with JellyFish against itself, and
see if one side continually wins more matches or more points than the other.
>All you've said here is that you're
>running a long series of unrealistically long matches
Hardly unrealistically long in that I want the luck factor reduced.
>and that is somehow
>supposed to prove something ill-defined about JF's rng.
Would you not agree that over the long haul "Jellyfish" and "Jelly2" should win
an equal number of 500-point matches and score roughly an equal number of
points?
"Jelly2" is the player's side (even though a bot is used), while "Jellyfish" is
the bot's.
>Isn't the whole deal you've postulated kind of meaningless in the short run
>of dice that marks most matches (even the 25 point matches)?
No more or less so than the casino's advantage over the player, which is
miniscule and won't impact whether or not most people win or lose on a given
night.
Hank Youngerman
matches would be more efficient.
Suppose you wanted to know if I'm better than Nack Ballard. So we
play a single 25,000-point match and he wins. This proves nothing, as
he could have won 25,000-24,986, and been a little lucky.
However, if we play 10,000 3-point matches and he beats me
5,500-4,500, this would be conclusive. This would represent 10
standard deviations and could not possibly be the result of luck.
For statistical purposes, you're much better off with 10,000 5-point
matches than with 100 500-point matches.
I'm not convinced of the value of the rest of the exercise, but if
you're going to do it, this method would be a significant improvement.
On 17 Jun 2002 00:04:15 GMT, amodern...@aol.com (A Modern Caveman)
wrote:
A Modern Caveman
>matches would be more efficient.
>
>Suppose you wanted to know if I'm better than Nack Ballard. So we
>play a single 25,000-point match and he wins. This proves nothing, as
>he could have won 25,000-24,986, and been a little lucky.
I don't know if this is true. With 500-point matches, you're going to get
fewer short-term fluctuations that win matches.
You could of course do both.
Cheating also is but one possible cause; it is more likely the result of the
random-number generator.
My main interest is in total points won (I'm tracking the score of each match),
and of course total matches won.
Peter Schneider
> I'm taking JellyFish on level 7 and having it play a series of 500-point
> matches against itself
> [...] This is an empirical way of seeing if the dice are balanced
> between the two bots (Jellyfish and Player2, who I call Jelly2).
Hmmm ... no matter how bad JF's prng would be -- since both program
instances are using it, your comparison is a dead race. Or am I missing
anything?
You may want to have a look at appendix E of JF 3.5's online help though for
further experiments.
Peter aka the juggler
A Modern Caveman
>> matches against itself
>> [...] This is an empirical way of seeing if the dice are balanced
>> between the two bots (Jellyfish and Player2, who I call Jelly2).
>
>Hmmm ... no matter how bad JF's prng would be -- since both program
>instances are using it, your comparison is a dead race. Or am I missing
>anything?
Yes, it SHOULD be a dead race. That doesn't mean it IS one, and if there is a
disparity, it doesn't mean it was intentional on the part of the programmer. I
doubt my commodore-64 BASIC programmers intended for their dice-generator to
spit out many more even than odd numbers either, but shit happens with
computers.
>
>You may want to have a look at appendix E of JF 3.5's online help though for
>further experiments.
I don't have 3.5 as far as I know.
So far, "Jellyfish" (the computer side) is leading "Jelly2" (the player's side)
by a match score of 31-29 (each a 500-point match), and is nailing 50.40
percent of the total game points, from a sample of about 47,000 games.
It's not a big disparity, but it seems to be holding up pretty steadily.
Whenever I'm bored I just run another series of matches to add to the running
total.
Kees van den Doel
A Modern Caveman <amodern...@aol.com> wrote:
>>> I'm taking JellyFish on level 7 and having it play a series of 500-point
>>> matches against itself
>>> [...] This is an empirical way of seeing if the dice are balanced
>>> between the two bots (Jellyfish and Player2, who I call Jelly2).
>So far, "Jellyfish" (the computer side) is leading "Jelly2" (the
>player's side) by a match score of 31-29 (each a 500-point match), and
>is nailing 50.40 percent of the total game points, from a sample of
>about 47,000 games.
>It's not a big disparity, but it seems to be holding up pretty steadily.
This is so incredibly STUPID.
>Ray Gordon, GENIUS
>http://www.cybersheet.com/library.html
Kees (The SubGenius Foundation, op Dirks dwizzy wrote: Jezus!)
A Modern Caveman
>
>This is so incredibly STUPID.
Is it?
Why would the computer bot defeat the player-bot for 50.40% percent of the
points?
That's a 0.8 percent advantage that should lead to winning about 51-52 percent
of the matches in what is supposed to be an EVEN distribution.
The number has bounced between 50.50 and 50.38 percent since the 30th match or
so.
The question is how many Elo points this would translate to.
Robert West
> >>It's not a big disparity, but it seems to be holding up pretty steadily.
> >
> >This is so incredibly STUPID.
>
> Is it?
>
> Why would the computer bot defeat the player-bot for 50.40% percent of the
> points?
>
> That's a 0.8 percent advantage that should lead to winning about 51-52 percent
> of the matches in what is supposed to be an EVEN distribution.
>
> The number has bounced between 50.50 and 50.38 percent since the 30th match or
> so.
>
> The question is how many Elo points this would translate to.
>
>
So next time you run the test from the beginning, and the player_bot
gets off to a very good start and the number bounces around between
50.50 and 50.38 for the player_bot, are you finally going to admit
that all of this proves nothing??
Biggles
Zorba
> >
> >This is so incredibly STUPID.
>
> Is it?
>
> Why would the computer bot defeat the player-bot for 50.40% percent of the
> points?
"Why not?" is the relevant question.
I know someone who tossed a coin 500 times as an experiment. It came
up heads 251 times!
Bob Lancaster
So, what is the statistical probability of this? How many standard
deviations from the expected 50:50 distribution?
As far as the match scores, 25:25 after 50 matches and 31:29 after 60
matches lie well within statistical fluctuations.
There are still several possibilities, listed with the most probable
first:
(1) Random distribution of luck.
It would be truly amazing if the results were exactly 50:50 for a
finite number of matches. Is 60 500 point matches really a large
enough sample to show a bias of 0.8%? I havn't done any statistical
analysis, but I would doubt the sample is large enough. (BTW, you are
making the claim, so it is your job to do the statistical analysis,
not mine.)
(2) Differences in cubing strategies when ahead or behind in a match.
There are at least some effects, even in a 500 point match, due to
cubing strategies when ahead or behind. This could create a
disproportionate number of match points won or lost within an
individual match. If I were behind 400-300 in a 500 point match, I
would probably be looser with the cube than if I were ahead at the
same score. I am not convinced that 60 matches is sufficient to
cancel out these effects. I also do not understand why you are using
matches of ANY length if you are counting total match points won or
lost. Wouldn't money play eliminate this possible bias?
(3) Differences in skill levels used under different circumstances.
It is at least theoretically possible that jf plays better when
rolling its own dice. Even a difference of a fraction of a percent
would be sufficient in this case.
(4) A small bias in the random dice generator that one instance of jf
has been able to exploit better than the other.
(5) Jellyfish cheats.
The question then is: how? Does jf roll get off the bar more often
than would be expected? Does jf use the Gaby Horowitz reverse
polarity magnetic dice? (High numbers in a race, low numbers in a
prime v. prime). Does jf figure out what would be the optimal roll,
and then increase the probability of that roll by a small percentage?
At this point, you have some interesting data. You have not shown
that you have enough data to be meaningful, nor have you shown that
you have correctly interpreted your data. At this point you seem to
assume (3) and (4) are incorrect, and you have finessed (1) and (2) to
go straight to number (5).
-Bob Lancaster
lancaster on FIBS
A Modern Caveman
>>
>> Why would the computer bot defeat the player-bot for 50.40% percent of the
>> points?
>>
>> That's a 0.8 percent advantage that should lead to winning about 51-52
>percent
>> of the matches in what is supposed to be an EVEN distribution.
>>
>> The number has bounced between 50.50 and 50.38 percent since the 30th match
>or
>> so.
>>
>> The question is how many Elo points this would translate to.
>>
>>
>
>So next time you run the test from the beginning, and the player_bot
>gets off to a very good start and the number bounces around between
>50.50 and 50.38 for the player_bot, are you finally going to admit
>that all of this proves nothing??
The player-bot DID get off to a good start, in fact.
I'm going to keep a long-term running total to see where it leads. Won't be
difficult to accumulate a few million points over time and see if any bias
sustains itself.
A Modern Caveman
>deviations from the expected 50:50 distribution?
That's what I'd like to know, as a backgammon "point" isn't a coinflip.
I do know that through 47,000+ points, the computer bot has secured 50.40
percent of the points.
>As far as the match scores, 25:25 after 50 matches and 31:29 after 60
>matches lie well within statistical fluctuations.
It was very close to being 32-28 or 33-27. These are 500-point matches.
Jelly2 won more of the close matches and still trails 31-29.
>There are still several possibilities, listed with the most probable
>first:
>
>(1) Random distribution of luck.
>It would be truly amazing if the results were exactly 50:50 for a
>finite number of matches. Is 60 500 point matches really a large
>enough sample to show a bias of 0.8%?
For those 60 matches. I intend to do many more. This is my first report.
>I havn't done any statistical
>analysis, but I would doubt the sample is large enough. (BTW, you are
>making the claim, so it is your job to do the statistical analysis,
>not mine.)
>
>(2) Differences in cubing strategies when ahead or behind in a match.
>There are at least some effects, even in a 500 point match, due to
>cubing strategies when ahead or behind. This could create a
>disproportionate number of match points won or lost within an
>individual match.
With no bias towards which side would win them.
> If I were behind 400-300 in a 500 point match, I
>would probably be looser with the cube than if I were ahead at the
>same score. I am not convinced that 60 matches is sufficient to
>cancel out these effects. I also do not understand why you are using
>matches of ANY length if you are counting total match points won or
>lost. Wouldn't money play eliminate this possible bias?
I can do other samplings. First I'm going to see how this holds up over a few
hundred or 1,000 matches of 500 points each.
>(3) Differences in skill levels used under different circumstances.
Shouldn't be a factor here.
>It is at least theoretically possible that jf plays better when
>rolling its own dice. Even a difference of a fraction of a percent
>would be sufficient in this case.
I'm not accusing JellyFish of anything. I'm trying to test the theory both
ways.
>(4) A small bias in the random dice generator that one instance of jf
>has been able to exploit better than the other.
The last time I toyed around with a "random" dice generator, the results were
horrid. I left the thing running for days and it was very obvious after a
short period of time.
>(5) Jellyfish cheats.
I'm not accusing the program of cheating.
>The question then is: how? Does jf roll get off the bar more often
>than would be expected? Does jf use the Gaby Horowitz reverse
>polarity magnetic dice? (High numbers in a race, low numbers in a
>prime v. prime). Does jf figure out what would be the optimal roll,
>and then increase the probability of that roll by a small percentage?
I'm not accusing anyone of cheating, I'm just collecting data.
The question about "cheating bots" wouldn't be HOW, but WHY they would have a
motivation to cheat. The only motive that comes to my mind is a very
impractical one and not even worth considering.
>At this point, you have some interesting data. You have not shown
>that you have enough data to be meaningful, nor have you shown that
>you have correctly interpreted your data. At this point you seem to
>assume (3) and (4) are incorrect, and you have finessed (1) and (2) to
>go straight to number (5).
I'm not accusing JellyFish or any bot of cheating. I'm just collecting data.
People go on and on about the bots so I figure it's better to do a real
experiment rather than just ranting and raving.
If the number holds at 50.40 percent for a million points I might say there's a
small bias, and then I'll run more simulations maybe with one-point matches to
see what happens there.
Peter Schneider
I said
> >
> >You may want to have a look at appendix E of JF 3.5's online help though
for
> >further experiments.
and you replied
> I don't have 3.5 as far as I know.
3.5 is the free version downloadable at
http://jelly.effect.no/download/jfl35.htm .
The help appendix E explains how to write your own pseudo random number
generator for JellyFish.
If (only) one of the JF instances would use a user-provided prng, there
might be a statistically significant difference in their probability to win
games. What you apparently do not understand is that your test can only
reveal a certain family of defects: those which give a different bias to the
rolls used by the JF instance which started, compared to the numbers for the
other JF instance; i.e. a defect that affects the 3rd and 4th, 7th snd 8th,
11th and 12th ... random number in a different way than the 5th and 6th, 9th
and 10th ... numbers (1). Some prngs may have such flaws, but to try to find
them by letting JF move the rolls at level 7 is hilarious. Let it at least
play it at level 1, you save two orders of magnitude CPU time! Or just sum
them up...
Btw, I don't see an obvious way to make only one JF instance use a different
prng because it must be coded as a *shared* library -- in particular, shared
by the two JF instances. (Are you at all aware of the fact that both JFs
probably use the same shared *code*, i.e. the physically same machine
instructions to create the rolls?) This would involve some hacking.
If you want to really test the quality of the Fish's prng, you should test
the numbers more efficiently than by letting JF play them (I'm still
laughing). Perhaps you can write a program that exploits JF's random number
generator API as explained in Appendix E (I suppose that the built-in prng
follows the same API) and generate a file with a huge amount of pr numbers
(very quickly! Your games have used only some millions of numbers so far --
takes 1 second on my laptop). These numbers should be tested for randomness
with standard tests, you may want to read http://www.fourmilab.ch/random/
where you can download one.
Really, try to understand why your experiment is actually not testing much,
and that little bit in the most ineffective way conceivable. If you manage
to perform a significant test on the Fish's dice though, let us know.
You could start with Snowie's dice -- there is an option to save 5000 rolls
to a file.
Peter aka the juggler
-------------
(1) Btw, even this should not affect the outcome of a multi game duel since
both players should get the starting roll roughly equally often and thus
would play with the "worse" dice equally often ... and would change "phase"
depending on the game length ... unless, coincidentally, the starting roll
itself would be biased, and in a very obvious way.
A Modern Caveman
score more than the other in ANY category long-term.
Is there a reason Jellyfish or Jelly2 should wind up with a long-term
percentage greater than the other, or more match wins? If you can come up with
one you would invalidate my experiment.
If you can't come up with this, then the only other option is to play out a few
million points worth and see if any biases hold.
Graham Price
"A Modern Caveman" <amodern...@aol.com> wrote in message
news:20020619023758...@mb-fk.aol.com...
> The premise of this experiment is simple: with NO BIASES, neither side
should
> score more than the other in ANY category long-term.
>
> Is there a reason Jellyfish or Jelly2 should wind up with a long-term
> percentage greater than the other, or more match wins? If you can come up
with
> one you would invalidate my experiment.
> http://www.cybersheet.com/library.html
maybe the obvious:
right now 31-29 is 50.4 and +2 match wins more
later on it maybe 30000100-29000900 50.00016 and +200 match wins more
Percentage gets closer but match wins is diverging.
Same holds true for flipping coins and there is no outside agency involved.
Zorba
> The premise of this experiment is simple: with NO BIASES, neither side should
> score more than the other in ANY category long-term.
Well, ANY category of infinity is (perhaps?) the right answer.
BTW, are you aware of the fact that if I perfectly toss a perfect coin
a million times, it is extremely unlikely it'll come up heads exactly
half the time? Some kind of (usually small) "bias" is much, much more
likely.
Zorba
> If the number holds at 50.40 percent for a million points I might say there's > a
> small bias, (...)
Don't you think it makes sense to find out *exactly* (i.e.
mathematically) if such an answer would mean
- a bias, and with what probability
- how big this bias might be and with which probability distribution
Otherwise your experiments will never tell you anything. Trying to
interpret statistics intuitively is asking for trouble.
Kees van den Doel
Zorba <zo...@chello.nl> wrote:
>> If the number holds at 50.40 percent for a million points I might say
>> there's a small bias, (...)
>Don't you think it makes sense to find out *exactly* (i.e.
>mathematically) if such an answer would mean
>- a bias, and with what probability
>- how big this bias might be and with which probability distribution
>Otherwise your experiments will never tell you anything. Trying to
>interpret statistics intuitively is asking for trouble.
This whole experiment is laughable of course, HA HAH HA HA HA!
For N trials standard deviation is sqrt(N)/2 if I remember correctly, so
for N = 30000 (assuming the GENIUS has played out 60 500 pt matches
which I think he said he did) a "normal" outcome would be
(15000+87)/30000 = 50.29%. 50.40% is close enough to show no indication
of any bias.
Of course anyone with at least 1 braincell knew this would come out.
Maybe the GENIUS will run some experiments to check if the earth is
REALLY round and not FLAT when he's finished with this excercise in
futility.
Kees (It must destroy the future, when I'll refrain.)
A Modern Caveman
>for N = 30000 (assuming the GENIUS has played out 60 500 pt matches
>which I think he said he did) a "normal" outcome would be
>(15000+87)/30000 = 50.29%. 50.40% is close enough to show no indication
>of any bias.
Then there isn't a bias.....I am drawing no conclusions.
A Modern Caveman
> right now 31-29 is 50.4 and +2 match wins more
Yes, but the "luck" factor (and total points) suggested that this is the
valley, not the peak. The peak was 50.50 and the valley was 50.38. The
numbers haven't changed much in the past twenty matches.
I would bet on Jellyfish at even money simply because if there is no bias I
have even-up odds anyway, and if there is I should win.
> later on it maybe 30000100-29000900 50.00016 and +200 match wins more
>Percentage gets closer but match wins is diverging.
>Same holds true for flipping coins and there is no outside agency involved.
But we haven't seen what happens over the next matches.
I'm sure everyone is awaiting this with bated breath.
A Modern Caveman
>should
>> score more than the other in ANY category long-term.
>
>Well, ANY category of infinity is (perhaps?) the right answer.
>
>BTW, are you aware of the fact that if I perfectly toss a perfect coin
>a million times, it is extremely unlikely it'll come up heads exactly
>half the time? Some kind of (usually small) "bias" is much, much more
>likely.
50.40 percent isn't proof, it's a start. If it snaps back, that will also be
evidence of no bias.
The experiment can go on forever!!
My C-64 computer was biased towards even numbers in its BASIC diceroller. No
explanation for it, but eventually the bias became clear.
Kees van den Doel
A Modern Caveman <amodern...@aol.com> wrote:
>>> The premise of this experiment is simple: with NO BIASES, neither
>>> side
>>should
>>> score more than the other in ANY category long-term.
>>Well, ANY category of infinity is (perhaps?) the right answer.
>>BTW, are you aware of the fact that if I perfectly toss a perfect coin
>>a million times, it is extremely unlikely it'll come up heads exactly
>>half the time? Some kind of (usually small) "bias" is much, much more
>>likely.
>50.40 percent isn't proof, it's a start. If it snaps back, that will also be
>evidence of no bias.
>The experiment can go on forever!!
>Ray Gordon, GENIUS
Be careful that your CPU doesn't overheat, Mr GENIUS.
Kees (When used on Monday morning cloudy periods.)
DarkSide
>
> 50.40 percent isn't proof, it's a start. If it snaps back, that will also be
> evidence of no bias.
>
Even if there is a bias in the random number generator, it very likely
will not show up in your experiment since both sides are using the
same RNG. Imagine an RNG that spits out only even dice rolls, would
you expect one side to win more than the other?
Do what you want, but I have to think there's better use of a genius's
time. For instance you could test Jellyfish's RNG for bias in a way
that would likely find the bias if there is one. Or (gasp!) you could
learn to play backgammon better. There are a number of good books out
there that can help you.
Also, I'm curious... have you figured out yet that some correct
doubles should be taken?
DS
A Modern Caveman
>be
>> evidence of no bias.
>>
>
>Even if there is a bias in the random number generator, it very likely
>will not show up in your experiment since both sides are using the
>same RNG. Imagine an RNG that spits out only even dice rolls, would
>you expect one side to win more than the other?
No, but this is exactly what's happening so far.
>Do what you want, but I have to think there's better use of a genius's
>time. For instance you could test Jellyfish's RNG for bias in a way
>that would likely find the bias if there is one. Or (gasp!) you could
>learn to play backgammon better.
Who says I don't play well?
>There are a number of good books out
>there that can help you.
>
>Also, I'm curious... have you figured out yet that some correct
>doubles should be taken?
Only as it relates to match play (like when the opponent is one point away from
winning the match), but not board play.
From a positional standpoint, it is either wrong to offer or wrong to accept
ANY given double. Has to be. Only the side with favorable chances would be
favored by increased stakes.
Otherwise you have a concept of "dual equity."
Gordon Parker ADVISORY
amodern...@aol.com (A Modern Caveman) wrote:
>
> >For N trials standard deviation is sqrt(N)/2 if I remember
correctly, so
> >for N = 30000 (assuming the GENIUS has played out 60 500
pt matches
> >which I think he said he did) a "normal" outcome
would be
> >(15000+87)/30000 = 50.29%. 50.40% is close enough to show no
indication
> >of any bias.
>
> Then there isn't a bias.....I am drawing no conclusions.
>
>
> Ray Gordon, GENIUS
(To avoid seeing this message please filter it via the name
"ADVISORY")
The previous post was made by "Ray Gordon", real name Gordon Roy
Parker.
Be aware this individual is known to be extremely disruptive in
the newsgroups he visits. He makes high volumes of posts
sometimes in excess of 60 posts in 24 hours. His actions include:
- being deliberately offensive, inflamatory, provocative and
antagonistic
- spamming Usenet with advertisements for his website and
products
- making accusations and claims of conspiracies, insisting he
has proof, but not providing any
- threatening people with lawsuits and litigation on an almost
DAILY BASIS. He has threatened people with legal action
literally
hundreds, perhaps thousands, of times. (In eight years he
only
actually sued one person from Usenet. While seeking
thousands
of dollars in judgement, he eventually accepted a $250
settlement offer)
- claiming he is in contact with numerous law enforcement
agencies
and that action against his detractors is imminent and will
be
"beyond their imagination". (Nothing has materialized after
8 years)
- contradicting himself and lying
- claiming he was Jesus in a past life
Some individuals have also reported being harassed by him
offline, with complaints being sent to the abuse department of
their ISP's or having their employers contacted by phone.
Mr Parker is known to be 35 years old and living with his mother.
* * *
Ray shows symptoms of the following illnesses:
- Narcissistic Personality Disorder
see diagnostic criteria :
http://www.psychologynet.org/npd.html
- Bi-polar Disorder (also called manic-depression)
see diagnostic criteria:
http://www.psychologynet.org/bipolar1.html
NOTE: Ray has stated he is indeed Bi-polar.
- Obsessive-Compulsive personality disorder
see diagnostic criteria:
http://www.psychologynet.org/ocdp.html
* * *
New readers are encouraged to familiarize themselves with Ray's
posting history, opinions and mental illness. Be sure to check
out www.Ray-Gordon.com and click the "RayFaq" button on the top
left. The site contains many of his archived posts. The site is
NOT affiliated with "Ray", who's real name is Gordon Roy Parker,
and he has made numerous threats of legal action against it.
* * *
Some other quotes from "Ray Gordon":
"There was no significant loss of life in those towers... NOT A
ONE"
Ray Gordon, real name: Gordon Roy Parker, September 11 2001
"A bunch of *ASSHOLE* New Yorkers died... don't grieve"
Ray Gordon, real name: Gordon Roy Parker, September 11 2001
And of course:
" ... I've warned people not to link to that site! "
Ray Gordon in reference to www.Ray-Gordon.com
* * *
Has "Ray" given you the impression that he runs a successful,
profitable business? Or that he commands expensive fees for his
work? Then you may be interested to know that he can't even
afford to pay his taxes. The IRS filed a tax lien of OVER TWELVE
THOUSAND DOLLARS against him and his Mom (whom he lives with)
which they STILL haven't been able to pay off AFTER SIX YEARS.
You can view the official document HERE: (cut and paste the
ENTIRE link onto a single line in your browser - will require
two separate pastes to get both lines onto single browser line)
http://dns2.phila.gov:8080/fjd/owa/zk_fjd_public_qry_03.zp_dktrp
t_frames?case_id=970620114
Gordon Parker ADVISORY
Gordon Parker ADVISORY
Gordon Parker ADVISORY
Gordon Parker ADVISORY
Gordon Parker ADVISORY
shorter
> >matches would be more efficient.
> >
> >Suppose you wanted to know if I'm better than Nack Ballard.
So we
> >play a single 25,000-point match and he wins. This proves
nothing, as
> >he could have won 25,000-24,986, and been a little lucky.
>
> I don't know if this is true. With 500-point matches, you're
going to get
> fewer short-term fluctuations that win matches.
>
> You could of course do both.
>
> Cheating also is but one possible cause; it is more likely the
result of the
> random-number generator.
>
> My main interest is in total points won (I'm tracking the
score of each match),
> and of course total matches won.
Gordon Parker ADVISORY
Gordon Parker ADVISORY
>
> I'm taking JellyFish on level 7 and having it play a series of
500-point
play out each time.
>
> I would advise others who are curious to run their own
simulations and see what
> results they get. This is an empirical way of seeing if the
> between the two bots (Jellyfish and Player2, who I call
Jelly2).
>
>
Gordon Parker ADVISORY
pretty steadily.
> >
>
>
> Why would the computer bot defeat the player-bot for 50.40%
percent of the
> points?
>
> That's a 0.8 percent advantage that should lead to winning
about 51-52 percent
> of the matches in what is supposed to be an EVEN distribution.
>
> The number has bounced between 50.50 and 50.38 percent since
the 30th match or
> so.
>
> The question is how many Elo points this would translate to.
>
>
Gordon Parker ADVISORY
don't factor
> >out the cube and playing skill and luck, you don't get a
clear picture of
> >the dice, do you? Why should points scored/matches won tell
you anything
> >about the dice? If Jf and Jf2 in my 1007 game sample were
exactly equal in
> >playing strength, and the dice were reversed, does that
result (assuming
> >it's the same) prove anything about the dice?
>
> 1,007 games isn't enough for what I'm doing here.
>
>
> >Is your focus on the dice, or on the Fish's play? If the
dice, then better
> >to print out 100,000 rolls of the dice from the rng and
analyze that
> >statistically, I would think. That tells you if the dice/rng
are good or
> >bad.
> >
> >Anyway, aside from all that, what's the starting point of the
exercise? Is
> >it that Jellyfish cheats?
>
> No. It's that IF there is a bias in the dice, intentional or
not, it will show
> up in a series of 500-point matches.
>
> A 500-point match really isn't luck, and total points won
should be roughly
> equal.
>
>
> >You didn't say that, but this kind of post usually
> >seems to spring from that well, and that well is dry.
> >
> >What does "balanced between the bots" mean? How do you define
that? Does it
> >mean each bot should get the same rolls?
>
> No. It means that there is no reason to expect either side to
win more often
> than the other.
>
> I remember when I had a commodore 64 that had a random-number
generator. I
> used it to make a dice simulator and after leaving it running
for days, was
> surprised to see the even numbers coming up much more often
than the odd ones.
>
>
> >The same number of certain classes
> >of rolls (doubles, non-doubles, ones, twos, threes,
etcetera)? If you want
> >to run an experiment, shouldn't you carefully define your
goals,
> >assumptions, terms, and methodology?
>
> Simple: play a series of 500-point matches with JellyFish
against itself, and
> see if one side continually wins more matches or more points
than the other.
>
>
> >All you've said here is that you're
> >running a long series of unrealistically long matches
>
> Hardly unrealistically long in that I want the luck factor
reduced.
>
>
> >and that is somehow
> >supposed to prove something ill-defined about JF's rng.
>
> Would you not agree that over the long haul "Jellyfish" and
"Jelly2" should win
> an equal number of 500-point matches and score roughly an
equal number of
> points?
>
> "Jelly2" is the player's side (even though a bot is used),
while "Jellyfish" is
> the bot's.
>
>
> >Isn't the whole deal you've postulated kind of meaningless in
the short run
> >of dice that marks most matches (even the 25 point matches)?
>
> No more or less so than the casino's advantage over the
player, which is
> miniscule and won't impact whether or not most people win or
lose on a given
> night.
Gordon Parker ADVISORY
Gordon Parker ADVISORY
Gordon Parker ADVISORY
Jellyfish. Took
> >about fifteen minutes. I don't see that the result tells me
anything about
> >the dice.
>
> I'm using 500-point matches. So far I've done 35 of them.
Zorba
> From a positional standpoint, it is either wrong to offer or wrong to accept
> ANY given double. Has to be. Only the side with favorable chances would be
> favored by increased stakes.
>
> Otherwise you have a concept of "dual equity."
I'll give it to you, GENIUS. You're better than Murat Kalinyaprak.
I'll remember this quote, good stuff to confuse potential fish!
--
><///_')
Frank Mazza
Enough with the "Gordon Parker Advisory," already.
He is not saying anything derogatory or inflammatory in these recent
posts, just talking about his experiment.
I don't agree with him about the merits of his experiment, but there
is nothing wrong with his performing it or discussing it. He doesn't
use profanity every other sentence like Murat.
Frank
Douglas Zare
Frank Mazza wrote:
> Enough with the "Gordon Parker Advisory," already.
>
> He is not saying anything derogatory or inflammatory in these recent
> posts, just talking about his experiment.
> [...]
Still, here I think he is clearly a troll, someone who intentionally
posts stupid statements in order to get a reaction. That's why he says
10*70 = 490 and signs it "Genius." I don't think he cares about his
experiment, and I don't think anyone else does.
The advisories are probably excessive, but it would have been nice to
see them earlier, when people may have thought he was simply confused
and misguided rather than willfully ignorant. At this point the
advisories are redundant.
Douglas Zare
Paul Tanenbaum
news:3D17D236...@math.columbia.edu...
> Frank Mazza wrote:
> > Enough with the "Gordon Parker Advisory," already.
> > He is not saying anything derogatory or inflammatory in these recent
> > posts, just talking about his experiment.
> > [...]
>
> see them earlier, when people may have thought he was simply confused
> and misguided rather than willfully ignorant. At this point the
> advisories are redundant.
The advisory is RepBot -
---
Paul T.
Murat Kalinyaprak
Zorba wrote 8e504b59.02062...@posting.google.com
> Caveman wrote 20020622230201...@mb-fl.aol.com
>> From a positional standpoint, it is either wrong to offer or wrong to
>> accept ANY given double. Has to be. Only the side with favorable
>> chances would be favored by increased stakes.
>> Otherwise you have a concept of "dual equity."
> I'll give it to you, GENIUS. You're better than Murat Kalinyaprak.
I knew you would dump me if someone better came along... :((
> I'll remember this quote, good stuff to confuse potential fish!
Are you confused...? Poor little fish... :((
If the same statement was made in a book or web site published
by one of your world-class fishermen, you would have swallowed
it line, hook and sinker... Poor little fish... :((
MK
-----= Posted via Newsfeed.Com, Uncensored Usenet News =-----
http://www.newsfeed.com - The #1 Newsgroup Service in the World!
-----== 100,000 Groups! - 19 Servers! - Unlimited Download! =-----
Lt. Tyrone Slothrop
> ANY given double. Has to be. Only the side with favorable chances would be
> favored by increased stakes.
>
> Otherwise you have a concept of "dual equity."
>
> Ray Gordon, GENIUS
Ray,
You are mistaken. Simple algebra:
Take a given position in a money game playing for a dollar a point
with a neutral cube and no chance of a gammon or backgammon by either
side. In this position, the probability of player#1 winning is 'P'.
Then the probability of player#2 winning is '1-P'. So the expectation
of the number of dollars player#1 would get more than player#2 (or in
other words, his 'settlement equity') is:
1P - 1(1-P) = 2P-1 dollars
If player#1 were to double in this position, and player#2 were to
take, then his settlement equity is:
2P - 2(1-P) = 4P-2 dollars
When is it better for player#1 to double then? Obviously when he
expects to get more money from player#2 than in the case he doesn't
double, or when:
4P-2 dollars > 2P-1 dollars
P > 0.5, which is just to say whenever player#1 is favored to
win.
However, player#2 may refuse a proffered double. In the previously
given position where the probability of winning is 'P', assume that
player#1 decides to double. If player#2 takes, then his settlement
equity is:
2(1-P) - 2P = 2-4P dollars
If however player#2 were to refuse the double, then he would always
lose 1 point compared to player#1. His settlement equity is:
-1 dollars
When is it better for player#2 to take a double than? Obviously when
he expects to get more points from player#1 than in the case he
refuses, or when:
2-4P dollars > -1 dollars
P < 0.75
1-P > 0.25, which is to say whenever player#2 has better than a
25% chance of winning.
So clearly if the given position should have a P of between 0.5 and
0.75 (say for example that player#1 has a 66% chance of winning), then
it would be completely correct for player#1 to double. But if
player#1 were to double, then nonetheless it would be completely
correct for player#2 to refuse.
This simple 'boundary value' case should be enough to show that there
is at least some art in the proper use of the cube, since it may be
the case that one or the other player has not correctly calculated the
value of 'P'. (And the math just gets more complex for cases of
matches, cube ownership, gammons and backgammons, etc.)
Christhecat01
think snowie is real which scans the board on everymove controls the dice rolls
only way u have a chance beating it is to understand what it will do
A Modern Caveman
In which case Player 1 should not double. You cannot have dual equity. Any
math which says you can is flawed on some level.
A bet can only be good for ONE side. That's Gambling 101. Only when match
score is taken into account is this not the case.
>or when:
> 2-4P dollars > -1 dollars
> P < 0.75
> 1-P > 0.25, which is to say whenever player#2 has better than a
>25% chance of winning.
You don't accept a double with a 25 percent chance of winning because the
GAMMONS you get creamed with are going to totally fuck up your results.
>So clearly if the given position should have a P of between 0.5 and
>0.75 (say for example that player#1 has a 66% chance of winning), then
>it would be completely correct for player#1 to double. But if
>player#1 were to double, then nonetheless it would be completely
>correct for player#2 to refuse.
My original point: a double can only be good for ONE side, except in a few
quirky positions where the match score is involved, or when both sides are
bearing off and gammons aren't possible.
>This simple 'boundary value' case should be enough to show that there
>is at least some art in the proper use of the cube, since it may be
>the case that one or the other player has not correctly calculated the
>value of 'P'. (And the math just gets more complex for cases of
>matches, cube ownership, gammons and backgammons, etc.)
Obviously.
I'm a converted chessplayer, and I've only been playing backgammon for about a
year (my strength is probably 1600-1800 if I had to guess), but I also gamble a
lot and have never seen dual equity in any form of betting.
David Montgomery
wrote:
>My original point: a double can only be good for ONE side, except in a few
>quirky positions where the match score is involved, or when both sides are
>bearing off and gammons aren't possible.
>
>
>Ray Gordon, GENIUS
I would like to thank Ray Gordon, GENIUS, for this superlative
exposition, which up til I had not understood.
Mr Gordon, GENIUS, I would like to direct your attention to
the poker newsgroups, where your talents are even more
urgently needed. Perhaps you are not aware, but sometimes
in poker, when one player makes a bet, the others do not
all fold!!! Although my understanding is not as clear as
your own, it does seem to me that this must involve some
kind of "dual equity", and I'm sure poker players could
profit from your participation.
David Montgomery
A Modern Caveman
>>quirky positions where the match score is involved, or when both sides are
>>bearing off and gammons aren't possible.
>>
>>I also gamble a lot and have never seen dual equity in any form of betting.
>>
>>Ray Gordon, GENIUS
>
>I would like to thank Ray Gordon, GENIUS, for this superlative
>exposition, which up til I had not understood.
>
>Mr Gordon, GENIUS, I would like to direct your attention to
>the poker newsgroups, where your talents are even more
>urgently needed. Perhaps you are not aware, but sometimes
>in poker, when one player makes a bet, the others do not
>all fold!!!
In THEORY, someone should always fold or not bet.
Dual equity doesn't exist.
>Although my understanding is not as clear as
>your own, it does seem to me that this must involve some
>kind of "dual equity", and I'm sure poker players could
>profit from your participation.
In poker, information is hidden. Not true of backgammon.
Please explain to me again how a bet of the same amount, on opposite sides of
an equation, can have a profitable expectation for both parties.
Graham Price
"A Modern Caveman" <amodern...@aol.com> wrote in message
news:20020718121758...@mb-mt.aol.com...
> In poker, information is hidden. Not true of backgammon.
> http://www.cybersheet.com/library.html
aww c'mon... do you really know what the next series of dice rolls
are going to be. Are future rolls open information?
Julian Hayward
Caveman <amodern...@aol.com> writes
>
>Please explain to me again how a bet of the same amount, on opposite sides of
>an equation, can have a profitable expectation for both parties.
A bet *in the situation you've described* doesn't. The point you seem to
be unable to grasp is, in backgammon the player for whom it's not
profitable cannot simply decline the bet (double) and carry on, there is
a penalty involved which is often worse than the unprofitable
expectation.
>Ray Gordon, GENIUS
>http://www.cybersheet.com/library.html
>
>
--
Julian Hayward 'Booles' on FIBS jul...@ratbag.demon.co.uk
+44-1480-210097 http://www.ratbag.demon.co.uk/
---------------------------------------------------------------------------
"A witty saying proves nothing" - Voltaire
---------------------------------------------------------------------------
Tom Keith
> From a positional standpoint, it is either wrong to offer or wrong to accept
> ANY given double. Has to be. Only the side with favorable chances would be
> favored by increased stakes.
You would think so, but it's not true. It is possible for a position to be
both a correct double and a correct take. The easiest way to see it is with
an example:
13 14 15 16 17 18 19 20 21 22 23 24
+------------------+ +------------------+ In the last stages of bearoff,
| | | X | both players have two checkers
| | | X | on their own two-point. The
| | | | cube is in middle. It is O's
| | | | turn to roll.
| | | O |
| | | O | Should O double?
+------------------+ +------------------+ Should X take?
12 11 10 9 8 7 6 5 4 3 2 1
The first thing to notice about this position is that the next roll determines
the game. If O bears off both checkers this roll, he wins immediately; if not,
his opponent will double (or redouble) and O will be forced to drop. There are
26 rolls out of 36 that win immediately for O (72%).
Here are the three possible cube actions.
(1) O does not double:
O wins 1 point 72% of the time = +.72
O loses 1 point 28% of the time = -.28
----
O's expected winnings = +.44
====
(2) O doubles and X takes:
O wins 2 points 72% of the time = +1.44
O loses 2 points 28% of the time = -.56
----
O's expected winnings = +.88
====
(3) O doubles and X drops:
O wins 1 point = +1.00
=====
So O does better if he doubles (+.88) than if he does not double (+.44).
And X does better if he takes the double (-.88) than if he drops it (-1.00).
The correct cube action for both sides: O doubles, X takes.
Tom Keith
Nis Jorgensen
wrote:
>
>You would think so, but it's not true. It is possible for a position to be
>both a correct double and a correct take. The easiest way to see it is with
>an example:
>
> 13 14 15 16 17 18 19 20 21 22 23 24
> +------------------+ +------------------+ In the last stages of bearoff,
> | | | X | both players have two checkers
> | | | X | on their own two-point. The
> | | | | cube is in middle. It is O's
> | | | | turn to roll.
> | | | O |
> | | | O | Should O double?
> +------------------+ +------------------+ Should X take?
> 12 11 10 9 8 7 6 5 4 3 2 1
>
>The first thing to notice about this position is that the next roll determines
>the game. If O bears off both checkers this roll, he wins immediately; if not,
>his opponent will double (or redouble) and O will be forced to drop. There are
>26 rolls out of 36 that win immediately for O (72%).
Such a simple example, and still wrong.
As your own calculations show, the situation after O rolls a 1 is a
double/take, with the expected winnings of X being 0.88 times the
value of the cube. Therefore, the relevant equities are:
(1) O does not double:
O wins 1 point 72% of the time = .72
O loses 0.88 point 28% of the time = -.25
----
O's expected winnings = .47
====
(2) O doubles and X takes:
O wins 2 points 72% of the time = 1.44
O loses 1.76 points 28% of the time = -.49
----
O's expected winnings = .95
====
(3) O doubles and X drops:
O wins 1 point = 1.00
=====
So it is still a double/take - but it's not the simplest possible
case.
--
Nis Jorgensen
Amsterdam
A coin has no memory.
Backgammon, however, is played with dice.
Kees van den Doel
A Modern Caveman <amodern...@aol.com> wrote:
>Please explain to me again how a bet of the same amount, on opposite sides of
>an equation, can have a profitable expectation for both parties.
You are right, it can't.
But there are two DIFFERENT bets here as the action of OFFERING a double
is different from the action of ACCEPTING a double, with different
choices involved.
The first choice is to either play for 1 point or for 2 points. The
second choice is to resign the game or play for 2 points.
Hence the "dual-equity", which is not dual at all, it's just two
different bets, which are related, but with their own characteristics.
I am looking forward to your next attempt at pretending to not
understand this, in which you undoubtedly will show great creativity
again.
Kees (Kees don't understand one about PRESTON's and generate an
internationalist perspective which am early instances where thou
cry?)
Tom Keith
> Such a simple example, and still wrong.
> As your own calculations show, the situation after O rolls a 1 is a
> double/take, with the expected winnings of X being 0.88 times the
> value of the cube.
Ah, yes. You're right. A better example would have been to put X's
checkers on his one point.
13 14 15 16 17 18 19 20 21 22 23 24
+------------------+ +------------------+
| | | X |
| | | X | The cube is in middle.
| | | | It is O's turn to roll.
| | | |
| | | O | Should O double?
| | | O | Should X take?
+------------------+ +------------------+
12 11 10 9 8 7 6 5 4 3 2 1
The next roll determines the game: If O bears off both checkers this
roll, he wins; if not, X wins. There are 26 rolls out of 36 that win
for O (72%).
Here are the three possible cube actions.
(1) O does not double:
O wins 1 point 72% of the time = +.72
O loses 1 point 28% of the time = -.28
----
O's expected winnings = +.44
====
(2) O doubles and X takes:
O wins 2 points 72% of the time = +1.44
O loses 2 points 28% of the time = -.56
----
O's expected winnings = +.88
====
(3) O doubles and X drops:
Paul Tanenbaum
> amodern...@aol.com (A Modern Caveman) wrote:
> >My original point: a double can only be good for ONE side, except in a few
> >quirky positions where the match score is involved, or when both sides are
> >bearing off and gammons aren't possible.
> >I also gamble a lot and have never seen dual equity in any form of betting.
>
You speak for us all.
> Mr Gordon, GENIUS, I would like to direct your attention to
> the poker newsgroups, where your talents are even more
> urgently needed. Perhaps you are not aware, but sometimes
> in poker, when one player makes a bet, the others do not
> all fold!!!
That is because there are many poker players who are merely gamblers,
who like to 'jack up' the pot for no reason. The skill in poker is
in the card play, betting has nothing to do with it. But some modern
gamesters, who think they have found some kind of 'improvements', have
introduced this notion of betting, calling, and folding, just to sell
their books to inhabitants of planet Moro.
> Although my understanding is not as clear as your own, it does seem
> to me that this must involve some kind of "dual equity", and I'm
> sure poker players could profit from your participation.
Indeed.
Not to mention, if MK and the Genius were to appear on rec.poker as a
tag team, it would probably peg the needle on my har dee har har meter.
---
Paul T.
A Modern Caveman
X takes because the increased equity of the double for O is less than the
decreased equity of a resignation by X.
Murat Kalinyaprak
Tanenbaum wrote c9bdbfd.02071...@posting.google.com
> That is because there are many poker players who are merely
> gamblers, who like to 'jack up' the pot for no reason. The skill
> in poker is in the card play, betting has nothing to do with it.
> But some modern gamesters, who think they have found some
> kind of 'improvements', have introduced this notion of betting,
> calling, and folding, just to sell their books to inhabitants of
> planet Moro.
I read the above with a smile on my face... :)) But it is kind of sad
because the joke is on you... :((
Poker players, errr..., should I have said "poker gamblers", don't
play with a "doubling" cube; which means they are not limited to
only doubling the stakes. They can raise the stakes by whatever
amount they desire...
Now, you all would certainly remember that not too long ago I had
suggested right here in this group, a variation of backgammon
that I had called Murat-gammon in which players could raise the
stakes by any number up to twice the current value of the "cube"...
For example, the first time around one can raise it from 1 to 2.
Then it can be raised to either 3 or 4. The next time around it can
be raised to 4, 5 or 6 or to 5, 6, 7 or 8 depending on whether it
was raised to 3 or to 4 the previous time, and so on...
When I had suggested this, a few people in here acknowledged
that it would make the hame much more complex but in a tone
of complaining about it as though it would be a bad thing...
If the ideal is to "improve" the game by making more complex
and by adding "greater levels of skill requirement", then what
was is the problem with my suggestion...?
I'm sure that if one of their world-class-bg-idols had suggested
it in a book or web site that they had paid for, the inhabitants of
planet Moro would have been happy to adopt it and incorporate
it by now at least in some bg servers or robots...
But the person who suggested it not being a world-class-gambler
and perhaps his name being "Murat" may be causing allergies to
the inhabitants of planet Moro...
If they don't like "Murat-gammon" they can call it "Kalin-gammon"...
But that may not work either since it kind of sounds like "Klingon",
who are the arch-enemies of the "Moron"s...
> Not to mention, if MK and the Genius were to appear on rec.poker
> as a tag team, it would probably peg the needle on my har dee har
> har meter.
At the time I had suggested "Murat-gammon", I wasn't particularly
thinking of poker... Why switch from our favorite "gambling game"
of "cubeful-backgammon" to poker...?? Why can't be play it just
like poker...???
For example, in a match of any given legth, with any current score
(i.e. any N-away, M-away bullshit), I would love the shove that "cube"
to somebody's face and say "I see your double and I raise the cube
to whatever you have in front of you on the table, in your wallet, in your
bank account, your wrist watch and your wife *bustard*...!!!
Heh heh hehhh... Despite my sometimes rude sounding tone here,
you guys can't even imagine how much I enjoy being among you in
this group...
I'm sure that people like Michael Crane who are trying to get rid of
their ugly wives would jump on a bet like the one above.... But "no
dice"... :)) And, BTW Michael, if you want to publish this article in
the next issue of "Biba-fax", you'll have to pay me a royalty... :)))
OK now, would any inhabitants of planet Moro want to play against
me a 25-point match of "Murat-gammon"...??? :)))
Lt. Tyrone Slothrop
> >When is it better for player#2 to take a double than? Obviously when
> >he expects to get more points from player#1 than in the case he
> >refuses,
>
> In which case Player 1 should not double. You cannot have dual equity. Any
> math which says you can is flawed on some level.
Ray,
This is not a case of "dual equity" since the decision by player#2 to
take or to refuse the double is actually *DIFFERENT* (and not
symmetrical) with the decision by player#1 to double.
If you don't understand this, then consider:
Imagine a variant of backgammon (let's call it "pseudo-backgammon")
where all the rules are the same, except that when one player offers
to double, the other player can refuse the double and the game just
continues until the end as if the double had never been offered in the
first place.
In such a game, your concept of "dual equity" would say that what is
good for one player (the doubling player) can not possibly be good for
the other player (the one deciding to take or refuse). Therefore if
one player correctly doubles, the other player should correctly
refuse. I would agree with this.
But compare the situation of the second player in this
pseudo-backgammon with the situation of the second player in regular
backgammon IN A POSITION WHICH THE FIRST PLAYER IS FAVORED TO WIN
(example: both players are bearing off but the first player has a
better chance of getting all of his checkers off first). In
pseudo-backgammon, the second player can refuse the double without
consequences -- the game just goes on. Then depending on the laws of
probability and the favor or disfavor of his gods, he gains or loses.
In regular backgammon, the second player has a situation which is "the
worse of both worlds". He can either take the double and since he is
favored to lose, he will likely lose twice as much, or he can drop and
AUTOMATICALLY lose the original bet! Unlike pseudo-backgammon, he
does not even have the option of unravelling the first player's double
since automatically losing a point is not a reverse of the original
doubling 'bet'.
Now wouldn't you say that the situation of the second player in
pseudo-backgammon is different than the situation in regular
backgammon? So why are you so surprised that the math for the
doubling decision in the case of regular backgammon is different than
the math for the doubling decision in the case of pseudo-backgammon?
If this is all still too abstract for you, consider the following very
concrete example with numbers large enough to make it real. Imagine
you are playing a game of regular backgammon for a $5,000 a point!
You are 'X' and the following position has been reached with a
centered cube:
+-------------------------------------------------------------+
|\------------------------+-----+----------------------------/|
|| 13 14 15 16 17 18 | | 19 20 21 22 23 24 || ||
|| | |XXX || X||
|| | |XXX || ||
|| | |XXX || ||
|| | |XXX || ||
|| | |XX || ||
|| | | || ||
|| | --- | || ||
|| || 1 || ||--|| O on
roll.
|| | --- | || ||
|| | | || ||
|| | BAR |OO || ||
|| | |OOO || ||
|| | |OOO || ||
|| | |OOO || ||
|| | |OOO || O||
|| 12 11 10 09 08 07 | | 06 05 04 03 02 01 || ||
|/------------------------+-----+----------------------------\|
+-------------------------------------------------------------+
In this simple apparently symmetrical (but pathological) bearoff
position, there is no possibility of a gammon or a backgammon since
both sides have gotten off one checker. However, nonetheless, clearly
'O' is favored to win since he is on the roll, although the
probability that 'O' wins versus 'X' wins is reasonably close --
approximately 59.8%:40.2%. By reasons of minimizing the statistical
variance, 'O' might not want to double, but consider what happens if
'O' does in fact offer the double. His expected winnings would be
higher (winning $10K instead of $5K), which would be born out if both
of you were to play 1,000,000 games from this exact position. In such
a series, 'O' could be expected to win $5,980,000,000 versus
$4,020,000,000 by 'X' if he were to double and 'X' were to take --
coming out ahead by $1,960,000,000. But if he doesn't double, then
over these 1,000,000 games, he would win $2,990,000,000 versus
$2,010,000,000 by 'X' -- coming out ahead by (only) $980,000,000.
Ok, so 'O' offers you, 'X', the double. But according to your theory,
if the double is "good" for 'O', then it's "bad" for 'X'. So you, as
'X', would drop and just lose the $5K...
Or would you? Would you really just accept the automatic loss of the
$5K by dropping, or would you instead decide to take a crack at the
$10K wager at close to 2:3 odds? Because if 'O' were to offer you the
double 1,000,000 times and you were to refuse 1,000,000 times, then
your opponent would win $5,000,000,000 and you would win exactly $0 --
he would come out ahead by $5,000,000,000. This is more than he would
have been expected to win from you ($1,960,000,000) if you had just
taken his double each time!
> A bet can only be good for ONE side. That's Gambling 101. Only when match
> score is taken into account is this not the case.
If player#1 makes a bet (through offering a double), then player#2
does not have a choice to "unbet" since to refuse the double is not to
unbet (because refusing results in the automatic loss of 1 point and
is therefore asymmetrical with doubling where player#1 doesn't
"automatically" lose anything if he offers the double whether or not
such an offer is a good idea).
That is why the inequality which expresses player#1's decision (4P-2
dollars > 2P-1 dollars) is different than the inequality which
expresses player#2's decision (2-4P dollars > -1 dollars). In my
example, the "-1 dollars" in the latter inequality is simply
expressing the result of dropping -- the automatic loss of a dollar.
>
>
> >or when:
>
> > 2-4P dollars > -1 dollars
> > P < 0.75
> > 1-P > 0.25, which is to say whenever player#2 has better than a
> >25% chance of winning.
>
> You don't accept a double with a 25 percent chance of winning because the
> GAMMONS you get creamed with are going to totally fuck up your results.
Obviously you missed what I said to introduce this example. I said,
and I quote my first sentence (with a highlight): "Take a given
position in a money game playing for a dollar a point with a neutral
cube and NO CHANCE OF A GAMMON OR BACKGAMMON BY EITHER SIDE."
I purposely used this assumption to simplify the mathematics. Now you
must grant in your experience of playing backgammon that there exist
positions where there is NO CHANCE of a gammon or backgammon, don't
you? After all, the example position I gave above meets this
criterion! If you only accept my argument only for such a position,
but not any other, it will be enough for me.
> >So clearly if the given position should have a P of between 0.5 and
> >0.75 (say for example that player#1 has a 66% chance of winning), then
> >it would be completely correct for player#1 to double. But if
> >player#1 were to double, then nonetheless it would be completely
> >correct for player#2 to refuse.
>
> My original point: a double can only be good for ONE side, except in a few
> quirky positions where the match score is involved, or when both sides are
> bearing off and gammons aren't possible.
Actually, it wasn't my point because I made a typo. I had meant "But
if player#1 were to double, then nonetheless it would be completely
correct for player#2 to *TAKE*." Sorry for the error. But this
corrected statement is completely consistent with the rest of what I
had written.
> >This simple 'boundary value' case should be enough to show that there
> >is at least some art in the proper use of the cube, since it may be
> >the case that one or the other player has not correctly calculated the
> >value of 'P'. (And the math just gets more complex for cases of
> >matches, cube ownership, gammons and backgammons, etc.)
>
> Obviously.
>
> I'm a converted chessplayer, and I've only been playing backgammon for about a
> year (my strength is probably 1600-1800 if I had to guess), but I also gamble a
> lot and have never seen dual equity in any form of betting.
Well I'm a chess player too. But I submit that this case has nothing
to do with either chess or backgammon, persay. It has to do with the
fact that the decision of one player to take a certain action is not
symmetrical with the decision of the other player to respond to that
action. Take any game where this is true, and we will find such an
example of what you have misleadingly labeled as "dual equity".
Lt. Tyrone Slothrop
> That is because there are many poker players who are merely gamblers,
> who like to 'jack up' the pot for no reason. The skill in poker is
> in the card play, betting has nothing to do with it. But some modern
> gamesters, who think they have found some kind of 'improvements', have
> introduced this notion of betting, calling, and folding, just to sell
> their books to inhabitants of planet Moro.
> Paul T.
Paul,
I assume that you understand that you are exaggerating here, since it
would be like saying the skill in the game of blackjack is in the card
play, betting has nothing to do with it. Try telling that to
card-counters. To perhaps oversimplify, in any game of chance which
consists of a series of rounds where the probability is winning or
losing changes from one round to the next, then it is always more
skillful to put more at risk (via appropriately scaled betting) when
you are more likely to win than when you are less likely to win. This
is because it doesn't matter in such a game, how many rounds that you
win. It matters only how much money you win or lose in total! I can
lose 99 out of a hundred rounds, but if on that one round, I should
bet the farm, I'd still come out ahead. Since estimation of the odds
is a nontrivial skill, for a player to understand when he is favored
to win and bet accordingly is also a skill.
Unfortunately, our opponents do learn, so if we all mechanically
raised our bets in the same way every time our estimation of the
probability we would win goes up, then our poker opponents would grow
to understand that and respond accordingly. So there is definitely
some science and some art (psychology) in betting in poker. Of course
this doesn't mean that everyone bets intelligently as you pointed out.
After all, there are still "gamblers" that play such
probabilistically "rigged" games as roulette, aren't there?
Zorba
> X takes because the increased equity of the double for O is less than the
> decreased equity of a resignation by X.
>
>
> Ray Gordon, GENIUS
Excellent, Mr. Troll^H^H^H^H^HGENIUS!
Now, please apply your concept of "dual equity" to this please, so
Murat Kalinyaprak (and Biggles and NIHILIST) can learn something about
the cube.
Jive Dadson
>
>
> A bet can only be good for ONE side. That's Gambling 101.
>
The reason it can be correct for player A to double and player B to take
(in a money game or match) is that it costs player B the current value
of the cube to drop. It's not as though I came up to you and offered a
proposition which was favorable to me, but which you could decline at no
cost.
Consider a draw poker game. There's whumpteen dollars in the pot.
(Whumpteen is a large number.) Your only opponent accidentally shows
you that he has a straight. You have a one card draw to a flush. Your
opponent bets ten cents. Do you fold? (If you do, you lose all
interest in that whumpteen dollar pot.)
Jive
P.s. If you meant your previous posts to be taken seriously, you are
nowhere near 1800, and very likely not near 1600. I played the game for
a living for four years, but my Games Grid rating is only 1841. (My
excuse is that I'm out of practice. Yeah, that's it.)
J.
Jive Dadson
> Now, you all would certainly remember that not too long ago I had
> suggested right here in this group, a variation of backgammon
> that I had called Murat-gammon in which players could raise the
> stakes by any number up to twice the current value of the "cube"...
Why the limit? In poker, that would be "half-pot limit".
Jive Dadson
>
> I'm a converted chessplayer, and I've only been playing backgammon for about a
> year (my strength is probably 1600-1800 if I had to guess) ...
I don't think you guessed very accurately.
Jive Dadson
>
>
> I'm a converted chessplayer, and I've only been playing backgammon for about a
I'm guessing you guessed wrong. (Jive - rating 1841 on Games Grid)
Ric
up to somewhere in the neighborhood of 4,000 points, another to twice that.
Not hard. Make bad plays, double when the bot has to take because my game is
so bad. Yadda yadda yadda. I must be a terrible person, treating a bot that
way.
Ric
P.S. I won both games when the bots got sulky over my mistreatment of them.
I made up for it by feeding them some extra memory chips.
"LostVegan" <NO.JUNKl...@hotmail.com> wrote in message
news:9ppsjusu7u3emk57t...@4ax.com...
> On Wed, 24 Jul 2002 02:31:10 -0600, "Murat Kalinyaprak"
> <mu...@compuplus.net> wrote:
>
> > For years I have been accused of "Jacking up" the cube against some
> >robots in long matches of "Jack-gammon" and I heve never been able to
> >get through the thick skulls of them Morons that "it takes two to tango"
and
> >"it takes two to jack up the cube"...
>
> It's easy to 'jack up' a cube in a match against a bot *if* your
> position is significantly inferior. I've done it myself against
> GGRaccoon. Not to skew the ratings system, but out of steaming
> frustration. The bots will obviously take bad (bad from the
> perspective that you should not have doubled them) doubles and throw
> the cube back at the next opportunity if the situation dictates. Two
> such cycles can bring the cube to 16 in a 9 point (max match length
> for GG bots) in just four moves. But the bot's only interest is
> equity, unlike his opponent who's doing stupid human tricks.
>
> So I would say that 'jacking' can only be attributed to the player who
> a) doubles at a clearly inappropriate point, or b) accepts a double
> when a drop is abundantly clear, so that he can inappropriately double
> (if the cube is still live) on his next turn.
>
> --
> Marty (to respond via email, remove NO.JUNK from email address)
>
> "to be yourself, in a world that tries, night and day, to make
> you just like everybody else - is to fight the greatest battle
> there ever is to fight, and never stop fighting" -- e.e. cummings
Murat Kalinyaprak
LostVegan wrote 9ppsjusu7u3emk57t...@4ax.com
> On Wed, 24 Jul 2002 02:31:10 -0600, "Murat Kalinyaprak" wrote:
>> For years I have been accused of "Jacking up" the cube against
>> some robots in long matches of "Jack-gammon" and I heve never
>> been able to get through the thick skulls of them Morons that "it
>> takes two to tango" and "it takes two to jack up the cube"...
> It's easy to 'jack up' a cube in a match against a bot *if* your
> position is significantly inferior...... The bots will obviously take
> bad doubles and throw the cube back at the next opportunity
> if the situation dictates...... But the bot's only interest is equity,
> unlike his opponent who's doing stupid human tricks.
Maybe you think so because you only tried this occasionally and
perhaps your "stupid human tricks" didn't help you win...
But suppose I'm playing a good number of long enough matches
(i.e. 20 times 25-point matches) against certain robots doing my
"stupid human tricks" consistently, persistently and end up winning
most of them...
Then what would you say...?
Do you guys agree that using the cube randomly in a large enough
number of matches will not alter the winning ratio between any two
given players of whatever strength...? If you do then let's continue...
Certain *rules* or skills are meaningful and useful only when both
sides abide by and play by them.
I may choose to defy the so-called "cube skill" and try to force the
cube usage in a match to be as close to random as possible...
My opponent obviously can't have any control over my cube actions
and the best he can do to keep playing his own game by doubling
or taking properly.
Since I will almost always double or redouble (except in rare 100%
situations), we know that all games will be played out and the cube
will be sitting on his side of the table at the end of almost all games.
If we assume that both sides are of equal stregth, the only thing that
will effect the final score would be how the matches each side win
coincide with what final cube values... I would venture to argue that
the difference after 4 billion games will be negligeable or at the most
small enough to belittle that so-called cube-skill as bulshit...
Because, cube ownership won't have a value for him or help him at
all since I will take and return each cube...
Dropping properly won't help him at all since each dropped cube will
be a pure present to me and since I will never drop, he will have to
win as many more played-out games as the games he will drop...
I can understand you peoples' being in deep denial. It took Morons
a very long time to accept that the Sun wasn't revolving around the
planet Moro also... It will be interesting to see who will be the first
ones to get their heads out of the sand...?
Murat Kalinyaprak
LostVegan wrote h06aku8kclv3nd4lq...@4ax.com
> On Mon, 29 Jul 2002 00:38:51 -0600, Murat Kalinyaprak wrote:
>>But suppose I'm playing a good number of long enough matches
>>(i.e. 20 times 25-point matches) against certain robots doing my
>>"stupid human tricks" consistently, persistently and end up winning
>>most of them...
> You can write a book-of-the-month and wind up on Oprah Winfrey.
> I'm sure that "Successful Cube Jacking -- How I Made The Bots
> Cry" will debut at #1 on the New York Times best seller list (fiction
> category, of course).
Does anybody remember the jokes about psychological experiments
in FIBS web pages and in this newsgroup...?
I propose another psychological experiment: I will write a book but as
the author's name I will use (with permission, of course) "Kit Woolsy"
or "Bill Clinton", err, I mean "Bill Roberty"... And see how many copies
it will sell or what kinds of reviews it will get, etc... :))
I'm smiling but it really saddens me that Earthlings are outnumbered
by the tourists from the planet Moro 99 to 1 on our own planet... :((
burper
experiment on fibs, only second to fibs itself: RepBot.
Almost a year online now and still collecting very useful
information on all manner of perverse human activity.
When it gets enough intelligence, it will publish a book
entititled "How I Made the Humans Cry".
Murat Kalinyaprak
burper wrote 3D49994...@yeahright.net
> I have proposed and implemented the greatest
> psychological experiment on fibs, only second
> to fibs itself:
:)) I like your putting it in these words... :))
> RepBot. Almost a year online now and still collecting
> very useful information on all manner of perverse
> human activity. When it gets enough intelligence, it will
> publish a book entititled "How I Made the Humans Cry".
I thought that RepBot kept track of droppers. What else
"useful" information does it collect...? But let's talk about
FIBS...
First: I always argued that the FIBS rating formula was at
best a fart-ass bullshit based on non-empirical, arbitrary
constant values...!!
Second: Even if FIBS had some sort of empirical ratings
formula, it would still be useless since nobody knows who
signs on with how many different accounts, etc...
Personally, during the past years, I promoted my own rating
of FIBS itself from "shit-pit" to "scum-pit"... :))
Third: What are "psychological experiments" are good for
unless we take some action based on the results...?? I, for
one, had suggested to establish a fund to help heal any sick
bastards among us but unfortunately none of the benevolent
saints of backgammondom has taken any interest in that... :((
As you stick your head in your ass, say: "Amen!"... It will help
you go to heaven...
Murat Kalinyaprak
LostVegan wrote 5khkku4q9bbgcefu7...@4ax.com
> On Fri, 2 Aug 2002 02:07:23 -0600, "Murat Kalinyaprak" wrote:
>> I, for one, had suggested to establish a fund to help heal any sick
>> bastards among us but unfortunately none of the benevolent
>> saints of backgammondom has taken any interest in that... :((
> First, heal thyself.
If you read my first article on this very subject, you will see that I was
not only including myself among the sick bastards in here but I was
also hoping to be one of the first beneficiaries of the "Heal the Sick
Bastards Fund"...
My admitting it validates the statistics that say there must be some
sick bastards in RGB. The same statistics would also suggest that
there must be some sick bastards within subgroups of RGB, such
as bg server operators, bg software developers, etc... So, we are
patiently waiting for a few of them to come out of the closet... :))