PC games
FERRANTE
Get Jellyfish.
Mark Ferrante
On Sun, 05 Jul 1998 15:19:06 GMT, va...@netcomuk.co.uk (Blakpawn)
wrote:
>Does anyone know of a really good backgammon game for my PC (windows
>95/pentium p2)?
>I have one, but I'm just too good for it!
>Elaine :-)
Home of THE SHEIK
Jason Berg
I work for Wyvern Studios, a game company in Utah. They are working on a
really cool backgammon game that will be playable over LANs, the
Internet, and modems. They are working on the AI phase now and would
love some expert help with roll-outs and the theory behind them. If
anyone would like to be a technical consultant or beta tester for this
game they should write to Russ Blau at bl...@wyvern.com. He is the
Quality Assurance lead there. We want to make a game that will have
great graphics and still appeal to the backgammon truists out there.
Thanks,
Jason
Murat Kalinyaprak
In article <35a0dbc3...@news.primenet.com> FERRANTE wrote:
>va...@netcomuk.co.uk (Blakpawn) wrote:
>>Does anyone know of a really good backgammon
>>game for my PC (windows 95/pentium p2)?
>>I have one, but I'm just too good for it!
>>Elaine :-)
>Get Jellyfish.
>
>Mark Ferrante
I would say don't. Especially to someone who
mentions she is already too good for another
game. If the downloadable trial version is a
measure of things, anyone good at backgammon
would probably get tired of Jellyfish pretty
fast. After a few matches, its cheating gets
so obvious that you may get a few smiles out
of it and even have a little more fun trying
to predict (quite successfully) what it will
roll next but that's about the extent of it.
After that, it ends up in the "recycle bin".
It might entertain beginners or intermediate
players long enough that it might be worth a
try though (at least as a learning tool)...
MK
Michael J. Zehr
> I would say don't [buy jellyfish]. Especially to someone who
> mentions she is already too good for another
> game. If the downloadable trial version is a
> measure of things, anyone good at backgammon
> would probably get tired of Jellyfish pretty
> fast. After a few matches, its cheating gets
> so obvious that you may get a few smiles out
> of it and even have a little more fun trying
> to predict (quite successfully) what it will
> roll next but that's about the extent of it.
> After that, it ends up in the "recycle bin".
> It might entertain beginners or intermediate
> players long enough that it might be worth a
> try though (at least as a learning tool)...
It's possible to predict Jellyfish rolls 100% of the time, not just
predict them "quite successfully." The mechanism to do this is the same
as the way to show that Jellyfish doesn't cheat, so it's not surprising
that anyone who can't predict the rolls 100% of the time might also
think Jellyfish cheats.
Here's how to accurately predict rolls (and also show that it doesn't
cheat) (and please, DON'T take my word for it -- try this yourself.
This works for jellyfish 3.0. You can test it on 2.0 to see if it works
for that too.):
Set up a position where Jellyfish has two checkers on your ace point and
six and seven checkers on its own 1 and 2 points. Give yourself a
5-point board with blots on your 7, 13, 16, 17 and 18. Go to "settings,
seed" and enter 15273 as the seed and 117 as the counter. Put Jellyfish
on roll holding a 2-cube. Start playing. It will roll a miracle 66,
hitting twice, and then you will roll a disaster 12, dancing on its
2-point board. It will then double you out -- obviously a cheat.
Now set up another position. Give Jellyfish a closed board and three
checkers on the bar. Put 6 checkers on your own 6pt and 5 checkers on
your own bar point. Put blots on your 1, 3, 4, and 5. Give yourself a
2-cube and have Jellyfish on roll. Go to "settings, seed" again and
enter numbers as above. Start playing. Jellyfish will roll a disaster
66, then you will roll a 12, which you can play 5-4 3-1. After the next
roll you can double Jellyfish out.
The sequence 66, 12 will happen 100% of the time when the seed and
counter start at 15273 and 117 respectively. This will happen whether
it's a miracle sequence for jellyfish or a miracle sequence for you.
-Michael J. Zehr
Dean Kezan
> Now set up another position. Give Jellyfish a closed board and three
> checkers on the bar. Put 6 checkers on your own 6pt and 5 checkers on
> your own bar point. Put blots on your 1, 3, 4, and 5. Give yourself a
> 2-cube and have Jellyfish on roll. Go to "settings, seed" again and
> enter numbers as above. Start playing. Jellyfish will roll a disaster
> 66, then you will roll a 12, which you can play 5-4 3-1. After the next
> roll you can double Jellyfish out.
>
> The sequence 66, 12 will happen 100% of the time when the seed and
> counter start at 15273 and 117 respectively. This will happen whether
> it's a miracle sequence for jellyfish or a miracle sequence for you.
>
> -Michael J. Zehr
That because Jelly Fish use fixed dices.
You can try to write any seed and any counter and you will find out that
not
only few dices are repeating but all.
Well this is only bad made dice generator,if you could proof that Jelly
Fish
use this forward dices,that would be really something.Until than try to
find
external dice generator or roll it by hand's.
P.S. If you are really interested for JF dices try in some really though
sitution (exp. JF need only 4-X to come in from bar and it get it
4-5)try to use second button from master dice,to repeat the rolls.
Some times it can be really unbelievable ( 4-4 , 4-5, 4-6 etc.)
Best regards -Dean Kezan
Rodrigo Andrade
No. The random number generator used isn't fixed. You seem to get many
repeated rolls because there are only 36 possible combinations from rolling
a pair of 6-faced dice. It's very likely that you'll see, say, the roll 64
more than once in a single game.
I'll do an experiment, run Jellyfish and check w/ me what you get. Open JF,
pull down the File Menu and click on New. Then Double click on Single game
in the window that pops up. Now pull down the Settings menu and choose
Seed... Set the values equal to the values I give below.
Seed: 26316
Counter: 0 --> note: the counter is incremented by 1 every time one of
the players roll the dice.
ME JF
61
61 64
63 62
43 21
See how you get the same values??? Now go back the File/New... and double
click on Single Game again (you can even close JF and re-run it). Now enter
the following values:
Seed: 58951
Counter: 0
ME JF
64
41 66
63 42
61 51
Again, you've got the same rolls. Try it one more time. This time the
counter is not zero:
Seed: 888
Counter: 45
ME JF
21
44 62
32 52
65 55
Still rolling the same numbers?? Sure you are. Also notice a couple of
things: on all 3 examples JF won the opening roll, and it'll keep winning it
for as long as you input these same values. Also, given that JF is always
playing on the same level (I mean, you didn't play with the Level menu...),
and you're always making the same moves, JF will always make the same moves
too.
That's what it's called a pseudo random number generator. It's not 100%
random (it's actually close to that), but it's the best we can get from
computers.
RODRIGO
Gary Wong
> JF's seed and counter scheme is in no way a proof of
> its not cheating. In fact, its knowing ahead of time
> all rolls to come is the problem itself. If one knew
> his opponent's every next roll, even average players
> could cream world champs. The question is whether JF
> makes use of this "pre-knowledge". Without access to
> its programmed code, one may not be able to prove it
> technically but may be able to offer arguments which
> could be proof to some extent practically.
If you are interested in this matter, I suggest you read question 10 of
Daniel Murphy's Mini-FAQ at http://www.cityraccoon.com/minifaq.html and
check up the references it contains. It would be quite simple to prove
that Jellyfish does cheat in the manner you suggest for a particular
position; all you would need to do is post an article like the
following:
"In the following board position [include a diagram here -- include cube
position, and score if a match game], Jellyfish [specify version and
level] makes [some particular move] when the seed and counter are
[whatever], but [a different move] when the seed and counter are
[something else]."
Given that nobody has found and published such an occurrence as long as
Jellyfish has been available, can you forgive my suspicions that perhaps
it _doesn't_ cheat like this -- maybe the reason that it frequently wins
is simply that it's just a better player?
If you are correct, and Jellyfish genuinely is a weak player with manual
dice, then I recommend you attend a tournament and announce that you
are willing to play against Jellyfish with manual dice for $10 a point
(or whatever). I'm sure there will be plenty of bystanders willing to
back Jellyfish at those stakes. If you are right, you could come away
very rich. Whether you are right or not, somebody would come away
very much wiser.
Cheers,
Gary.
--
Gary Wong, Department of Computer Science, University of Arizona
ga...@cs.arizona.edu http://www.cs.arizona.edu/~gary/
Murat Kalinyaprak
> Murat Kalinyaprak wrote:
>> ..... After a few matches, its cheating gets
>> so obvious that you may get a few smiles out
>> of it and even have a little more fun trying
>> to predict (quite successfully) what it will
>> roll next but that's about the extent of it.
>It's possible to predict Jellyfish rolls 100% of the time, not
>just predict them "quite successfully." The mechanism to do
>this is the same as the way to show that Jellyfish doesn't
>cheat, so it's not surprising that anyone who can't predict
>the rolls 100% of the time might also think Jellyfish cheats.
It looks like we understand "predicting" differently
and what I meant may become more clear to you and to
other readers a little further down.
I started reading this newsgroup recently and I have
already read some past articles, including a few old
ones about Jellyfish's cheating, etc. At this point,
I feel that people defending JF on this subject must
be either taking a cut from its revenues, or must be
a little naive.
JF's seed and counter scheme is in no way a proof of
its not cheating. In fact, its knowing ahead of time
all rolls to come is the problem itself. If one knew
his opponent's every next roll, even average players
could cream world champs. The question is whether JF
makes use of this "pre-knowledge". Without access to
its programmed code, one may not be able to prove it
technically but may be able to offer arguments which
could be proof to some extent practically.
I have read articles where some writers were arguing
that JF's seeming luck is due to its positioning its
pieces in such a way to make better usage out of the
next roll (no matter what it rolls). Of course, this
is a naive argument in that it assumes the other guy
is picking berries in the hills, while JF is placing
its pieces in optimum spots as it wishes. It totally
ignores the fact that, the other player is trying to
do the very same thing also.
The difference in playing against JF letting it roll
the dice vs. rolling one's own dice is as obvious as
day and night. Some days ago I had posted an article
on this subject (which didn't seem to appear on this
newsgroup), relating the results of the very first 3
matches I played against JF (level 7-100) rolling my
own dice. I'll repost it here:
ME JF
-- --
25 6 (from 13-6)
40 6 (from 8-6)
20 0 (from 4-0)
Had I all of a sudden become a much better player or
was it pure luck 3 times in a row...? The day before
I won 2 matches against JF only hours apart, both of
which ended as 32-0 single game matches. Can anybody
in this newsgroup tell me when was it the last time
they beat JF like this when JF was rolling the dice?
I doubt anybody could. I don't think it would happen
while JF is rolling the dice. Today I beat JF 72-4,
from 8-4. The last one was a 32 point game and I did
gammon. Notice that JF lost by a gammon after having
doubled me all the way to 32! It was so obvious that
I couldn't help doubling back each time. How could a
smart player like JF not know that it was being very
badly gammoned but kept doubling instead? Perhaps it
is not refined enough to know the difference between
who is rolling the dice and keeps positioning itself
and waiting in vain for that "next" roll...?
Obviously once the cube got passed the 15-4 mark, it
didn't matter how high it went and JF kept doubling.
Is this the computer program that's supposed to be a
good tutor? It may be good gambling teacher, but I'm
not sure if it could teach graceful backgammon... Or
is it that backgammon is seen as no more than a tool
for gambling anymore?
>...[shortened]... The sequence 66, 12 will happen 100% of
>the time when the seed and counter start at 15273 and 117
>respectively. This will happen whether it's a miracle
>sequence for jellyfish or a miracle sequence for you.
The simple question is whether such sequences occur
more often than the odds. I don't care about seeds,
counters, etc. Since I can't possibly know/memorize
all those millions of seed/counter combinations, my
chances of guessing whether the next roll will have
a certain number per dice is 1 in 6. If I'm able to
guess better than that, either I must be psychic or
I must be observing something tangible, which helps
me accomplish this.
Here is a simple and practical challence that may be
put to people who think or argue JF does not cheat:
(I'm not much of a gambler and "I" will be used here
only figuratively to mean anyone who shares my view)
both put their money on the table, and I will start
playing against JF while the other watches. Whenever
I feel like it (before any roll) I'll say "Look Joe,
JF will now roll a 5 (or a 4-3, or not a 2, etc.)".
With my odds of being right 1 out of 6/being wrong 5
out of 6 (per dice), I will bet 1 dollar against the
other's 5 dollar (per dice). Enough games/rolls will
be played to let it even out (as it's supposed to do
so at least in theory). If JF indeed does not cheat,
neither one will get hurt (i.e. loose or win money).
But if I see JF playing in certain ways to give me a
hint that it might be relying on the next roll, then
whoever is betting against me may be in trouble. And
I get the impression that many other people have the
same observation as I do and perhaps they could also
do just as good at guessing JF's next roll.
Furthermore, I read all kinds of baloney about neural
and fuzzy logic stuff, as explanations for why JF not
always makes the "best" moves. Perhaps a much simpler
explanation may be that a "not so perfect" move suits
the "next roll" better...?
Let's keep in mind that JF can play at various levels
and that means it knows how to not make the best move
all the time (i.e. negative cheating to play at lower
levels) and let its opponent to win "proportionately"
also. Now, imagine JF can play at level 8, and at the
highest level of 7 that is currently selectable, it's
just "allowing" you to win enough to make you believe
that it's one notch above you and to keep you going.
Of course, this in itself may be argued to be a great
accomplishment on the part of JF. But, like I said at
the beginning of this article, JF itself may not have
a prayer against even a beginner, if that player knew
every next roll while JF didn't...
MK
Murat Kalinyaprak
>P.S. If you are really interested for JF dices try in some
>really though sitution (exp. JF need only 4-X to come in
>from bar and it get it 4-5)try to use second button from
>master dice,to repeat the rolls. Some times it can be
>really unbelievable ( 4-4 , 4-5, 4-6 etc.)
Just in case it was misunderstood, I wasn't proposing
that JF gets itself out of "tough positions", etc. by
rolling what it needs. JF's seed/counter scheme takes
care of any such claims already. What I was proposing
was that, it "leads" the game (and also the opponent)
to that position.
I have a feeling that JF also relies on its opponents
to make the moves expected from them, at the selected
difficulty level. When JF is rolling the dice, it may
perhaps be thrown off by a purposefully made bad move
temporarily but the effect may not last long since it
may/will make its following move based on to the next
roll to come again.
Speculating and experimenting with such things is fun
and perhaps the developers of JF may be having a ball
reading what's written on this subject in this group.
What seems to be sufficiently clear to a lot of us is
that, JF plays much better when it rolls the dice and
not all that good when the dice is rolled manually...
MK
John Goodwin
Kalinyaprak) wrote:
>Just in case it was misunderstood, I wasn't proposing
>that JF gets itself out of "tough positions", etc. by
>rolling what it needs. JF's seed/counter scheme takes
>care of any such claims already. What I was proposing
>was that, it "leads" the game (and also the opponent)
>to that position.
>
>I have a feeling that JF also relies on its opponents
>to make the moves expected from them, at the selected
>difficulty level. When JF is rolling the dice, it may
>perhaps be thrown off by a purposefully made bad move
>temporarily but the effect may not last long since it
>may/will make its following move based on to the next
>roll to come again.
>
>Speculating and experimenting with such things is fun
>and perhaps the developers of JF may be having a ball
>reading what's written on this subject in this group.
>
>What seems to be sufficiently clear to a lot of us is
>that, JF plays much better when it rolls the dice and
>not all that good when the dice is rolled manually...
This seems to be a good time to re-post the instructions for proving
that JF does not cheat, either by fixing the throws, or using
foreknowledge of the dice.
Repost
------------------------------------------------------------
It is fairly common for someone to post a question here
(rec.games.backgammon), querying the apparent good luck Jellyfish (JF)
has with its dice.
I have prepared this step by step guide for anyone who is not
convinced by the simple reassurances of those who have used and trust
it.
Section A
This is for those of a seriously pedantic disposition
You cannot *prove* that JF *never* manipulates the dice (at least
not in any feasible time span).
You cannot prove absolutely, that JF did not manipulate the dice
for a certain game, because there is an absurdly small possibility
that it is designed to win some specific games.
Neither of these caveats should concern anyone who does not work for
the security services of a disturbingly paranoid government
Section B
To satisfy yourself that JF has not manipulated the dice in any
particular game.
At the end of the game you wish to examine, note down the seed and
counter. They are to be found in the status bar, at the bottom right
of the JF window.
Print out the moves of the game. These can be found in the file
GAME.GAM, in the directory from which JF is run.
Count the total number of moves for each player, and subtract this
from the count that you have noted.
XX:
Start a new game, and, at some point, go to Settings|Seed, and set
the seed and counter to the values they were at the start of the
match you are checking.
Continue to play. You will note that the dice that are generated are
exactly the same as those from the match that you have printed.
You may repeat the steps from XX as many times as you like, and you
will see that no matter what the state of the board, JF always
generates the same sequence from the given seed and counter.
Section C
To satisfy yourself that JF is not using foreknowledge of the dice it
is about to generate.
Play a game, and produce a listing of GAME.GAM.
Check Settings|Manual Dice, so that you can feed in the dice
throws yourself.
Now, JF will have no way of knowing what dice are about to be
thrown.
Play the game, feeding in the dice values from the game you have
listed.
You will see that it proceeds in exactly the same way as it did when
JF was generating the dice.
This demonstrates that JF will play the same board, the same way,
even when it is not generating the dice, and thus can have no idea
what throws will be made.
Section D
It has been pointed out that JF could (in theory) be cheating, and
keeping a record of its activities so that if you try to catch it out
if would behave as if it was innocent of any wrong doing.
If you are *that* paranoid, then you might care to run the tests on a
different machine to that on which the initial games were played.
Make sure that you disconnect both machines from any kind of network
first (before running the initial game).
That might convince you, but some good quality psychiatric treatment
would probably be more beneficial.
---------
Do not take the contents of section D to imply any suggestion that you
are being unreasonable in questioning JF in the first instance.
If you have never played intermediate to strong players before, and
are unaware of the solid reputation that JF has amongst the many good
players who use it regularly (and the not so good ones such as
myself), you would need to be somewhat gullible not to seek some
reassurance.
JG
my_...@hotmail.com
of the revenues from jellyfish.
In article <6ojm06$nbr$1...@news.chatlink.com>,
mu...@cyberport.net (Murat Kalinyaprak) wrote:
> JF's seed and counter scheme is in no way a proof of
> its not cheating.
It is impossible to prove that Jellyfish does NOT cheat.
It is possible to prove that Jellyfish DOES cheat, however
no one has ever done so, though many have tried, hence the
widely held view that Jellyfish doesn't cheat.
Two things you should know, for your interest, not that
these facts "prove" anything. One, Jellyfish (ver. 3) playing
under a completely different random number generator on FIBS
(first internet backgammon server), attained a higher
average rating than any human player. And there are lots
of strong players on the server.
Two, (to my knowledge) no reputible or highly ranked player has
ever accused Jellyfish of cheating. Many reputible/ highly
ranked players use jellyfish as an analysis tool.
But yes, there are a lot of doubters out there.
> ME JF
> -- --
> 25 6 (from 13-6)
> 40 6 (from 8-6)
> 20 0 (from 4-0)
>
It would be more useful if you mentioned the length of
the match.
>
> I doubt anybody could. I don't think it would happen
> while JF is rolling the dice. Today I beat JF 72-4,
> from 8-4. The last one was a 32 point game and I did
> gammon. Notice that JF lost by a gammon after having
> doubled me all the way to 32!
That's interesting. I imagine it's possible (though very
unlikely) that your jellyfish is buggy. You didn't mention,
did you buy the game or download the free trial version?
At any rate, we can figure out the problem if you record
your matches (jellyfish can save the match to a text file)
and post them to this newsgroup. That way: other jellyfish
users can enter the same rolls to see if their copy of the
program makes the same moves, and we can see these positions
in which jellyfish is apparently erroneously running up
the cube. These kinds of posts are appreciated in this newsgroup
because they can be very enlightening and help to de-mystify
your arguments.
Also, if it is your belief that jellyfish cheats, you must
also think that it is making incorrect moves. Can you post
examples of what you consider a Jellyfish error? (and include
the match length and score at the time)
-----== Posted via Deja News, The Leader in Internet Discussion ==-----
http://www.dejanews.com/rg_mkgrp.xp Create Your Own Free Member Forum
Patti Beadles
Murat Kalinyaprak <mu...@cyberport.net> wrote:
>I have read articles where some writers were arguing
>that JF's seeming luck is due to its positioning its
>pieces in such a way to make better usage out of the
>next roll (no matter what it rolls). Of course, this
>is a naive argument in that it assumes the other guy
>is picking berries in the hills, while JF is placing
>its pieces in optimum spots as it wishes. It totally
>ignores the fact that, the other player is trying to
>do the very same thing also.
And the fact is, Jellyfish is *much* better at it than
anybody who's been complaining here.
Jellyfish is better than (pulling a number out of the air) 99%
of the backgammon players in the world. And I've not yet seen
a single world-class player complain that JF cheats. In fact,
every strong player I've heard speak about the subject has
defended the program's honesty.
-Patti
--
Patti Beadles |
pat...@netcom.com/pat...@gammon.com |
http://www.gammon.com/ | If it wasn't for the last minute
or just yell, "Hey, Patti!" | I'd never get anything done!
Vince Mounts
and got lucky 3 times. Would be curious to see long-term results with this
method.
And as another poster wrote. If you are that good at beating J/F then you
should find someone to back it for money. I would be more that happy to back
J/F at $10/point. Perhaps playing through FIBS and me entering the rolls to
J/F at level 7-100. Let me know if you are interested.
-----
Vince Mounts (a.k.a einniv)
E-Mail: vmo...@mindspring.com
Home Page URL: http://vmounts.home.mindspring.com
Murat Kalinyaprak wrote in message <6ojm06$nbr$1...@news.chatlink.com>...
Robert-Jan Veldhuizen
JE> On Thu, 16 Jul 1998 02:13:11 GMT, mu...@cyberport.net (Murat
JE> Kalinyaprak) wrote:
>>Just in case it was misunderstood, I wasn't proposing
>>that JF gets itself out of "tough positions", etc. by
>>rolling what it needs. JF's seed/counter scheme takes
>>care of any such claims already. What I was proposing
>>was that, it "leads" the game (and also the opponent)
>>to that position.
JE> <----------------- perfectly blocked ---------------->
JE> Murat does it again! It must be deliberate, surely?
Quite an achievement, I'd say.
It's not so easy to do, as I'm
finding out myself at this ver
y moment! Maybe his articles a
re computer-generated? Or mayb
e it is a subtle indication of
the troll-value of his posts??
(:--------------------------:)
--
Zorba/Robert-Jan
Rew Francis
jeib...@revolver.nomed.co.uk (James Eibisch) wrote:
>On Thu, 16 Jul 1998 01:51:41 GMT, mu...@cyberport.net (Murat
>Kalinyaprak) wrote:
>
>>I started reading this newsgroup recently and I have
>>already read some past articles, including a few old
>>ones about Jellyfish's cheating, etc. At this point,
>>I feel that people defending JF on this subject must
>>be either taking a cut from its revenues, or must be
>>a little naive.
>
>[snip]
>
>I'll leave it to others to argue the case for fairness in JF's dice.
>My interest is more typographical.
>
>In Murat's post there are 90 mid-paragraph, non-terminating lines
>(e.g. five in the above paragraph). Only four of them don't left- and
>right-align perfectly within the paragraph, and those four are off by
>only a single character. Nearly the whole post is perfectly blocked
>(using any fixed-width typeface), without the use of extra padding
>with spaces (remember character-based wp programs and their ugly text
>justification?).
>
>Is this by design, or by rather freaky coincidence? Perhaps there's
>something magical about a 52-character column width, but when I set
>some text of my own to 52 columns it was highly ragged, as is this
>reply I'm writing here. I'm intrigued, if not also rather sad for
>noticing this!
>
>Now what else can I do to put off starting this work I've got to
>finish before tomorrow? :-)
>
>Apologies for the off-topicality.
>
>--
>James _ To mail me, spell "nomed" in my address backwards
>Eibisch, ('v')
>'Ivan' (,_,) N : E : T : A : D : E : L : I : C : A
>on FIBS. ======= http://www.revolver.demon.co.uk
James, I think you could be on to something here.
The chances of something like this occurring in a
post of reasonable length which has been composed
without any thought being given to the phenomenon
you mention have to be truly astronomical, around
7.88860905221*10^69 to 1 according to my estimate
which I reached with a bit of back-of-an-envelope
scribbling. Assuming that there's an average word
length of five letters in the article (this comes
from the the usual way of calculating your typing
speed in words per minute given that your rate of
keystrokes is known - a reasonable assumption for
getting a ball-park figure) then one would expect
a line to end exactly on a given character length
about one-fifth of the time. Now, Murat's article
is about a hundred lines long. Given that the top
line is 52 (or whatever) characters in length, it
seems to me that the chances of a particular line
elsewhere sharing this identical length are going
to be about 1/5. And the chances of all the other
one hundred lines having that same length will be
1/5 to the power 100, i.e. the number I mentioned
above. Still, strange things happen. For instance
I saw recently reported in the English press that
a woman in West London had sliced a tomato in two
only to discover that it had an important passage
from the Koran visible in the pattern of the pips.
Rew
Rodrigo Andrade
>a single world-class player complain that JF cheats. In fact,
>every strong player I've heard speak about the subject has
>defended the program's honesty.
We never will. These guys have a lot of credibility and they could take a
company out of business if they started saying that a program cheats. What
would happen to Nike if Michael Jordan said their shoes are uncomfortable,
and just plain suck???
RODRIGO
bj...@lehigh.edu
Rodrigo, please give it up.
Several things I've noticed -- 1.) You cannot possibly advocate buying Snowie
or Jellyfish because it costs too much. Yet, you wholly endorse GG. Not
playing GG for 4 years is an equitable exchange for one of these programs.
2.) Someone posted a "lessons from the board" problem, stating something to
the effect that this newsgroup should try to return to its backgammon roots.
You posted another message in response, agreeing, stating "That's what this
newsgroup is all about...", yet you continue to post repetitive, useless
drivel.
3.) You refute the experts' endorsement of the neural nets. Your rating on
FIBS is in the mid-1300's. I know nothing about your rating on GG, so this
isn't the strongest argument.
4.) You frequently skip over any challenges of your arguments, and continue to
post, as if they've never existed.
5.) You respond without reposting the original message, obfuscating the
r.g.b. audience.
You have every right to post. However, your continual, fragmented, redundant
cries against FIBS, neural nets, and the pricing of backgammon products had
lost its novelty a long time back. Either say something that benefits the
group, or let others have a try at it.
Go to www.dejanews.com, and run a search on Kit Woolsey and r.g.b.. Read some
of his posts. Try running your own name and r.g.b., and compare the two.
Brad Mampe
bj...@lehigh.edu
Rodrigo Andrade
>or Jellyfish because it costs too much. Yet, you wholly endorse GG. Not
>playing GG for 4 years is an equitable exchange for one of these programs.
Just because I complain about the pricing, doesn't mean that I cannot buy
it. I'll not buy anything I think is overpriced just because I've got the
money.
>2.) Someone posted a "lessons from the board" problem, stating something
>to
>the effect that this newsgroup should try to return to its backgammon
roots.
>You posted another message in response, agreeing, stating "That's what this
>newsgroup is all about...", yet you continue to post repetitive, useless
>drivel.
That was Michael Zher. I support that idea, but I don't think we should be
strictly limited to that.
>3.) You refute the experts' endorsement of the neural nets. Your rating on
>FIBS is in the mid-1300's. I know nothing about your rating on GG, so this
>isn't the strongest argument.
Actually it's in the high-1300 :-) If you're smart enough to find my whois
info, you must also be smart enough to find out when the last time I logged
in there was. Also look at my ridiculously low experience. I haven't
actually played a game there in ages. Last time I logged in was about a week
ago, just to try to meet my friend on-line and get him started.
>4.) You frequently skip over any challenges of your arguments, and continue
to
>post, as if they've never existed.
Why am I replying this post, then??
>5.) You respond without reposting the original message, obfuscating the
>r.g.b. audience.
Do you know what the sign > mean on a newsgroup???
>Go to www.dejanews.com, and run a search on Kit Woolsey and r.g.b.. Read
some
>of his posts. Try running your own name and r.g.b., and compare the two.
Better yet: I ran "Brad Mampe" and look what I found: 8 out of 18 matches to
rec.games.backgammon (the other 10 to rec.games.bridge, if anyone's
wondering...). Also some of these posts actually belong to people who quoted
him. Now, let's examine some of these 8 bites of wisdom:
Subject: Cheats (was Re:my last word)
From: bj...@Lehigh.EDU
Date: 1998/06/30
Message-ID: <6namum$n...@ns4-1.CC.Lehigh.EDU>
Newsgroups: rec.games.backgammon
[Subscribe to rec.games.backgammon]
JG wrote:
[snip long unnecessary quote]
I know exactly how you feel. And I've got proof that JF cheats.
Last night, I opened 6/5, 24/21 off of 3-1. JF rolled 4-4. I danced, JF made
the five point, and I danced again. JF then rolled 5-5, pointed the 1, and
went on to win a backgammon.
After my opening roll, I never got to move!!!! And every time the opening
rolls go 3-1, 4-4, I can't ever win!!!!! JF cheat$ !
Brad Mampe
bj...@lehigh.edu
HEEEEEEEEEEEEEEEEEEEEY!!!!!!!!!!!!!!! It seems like I'm not the only one
questioning the opinion that JF cheats, eh??? Let's move on:
Subject: Re: Cheats (was Re:my last word)
From: "Harold Evenson" <Harold....@v-wave.com>
Date: 1998/06/30
Message-ID: <6nbhkd$q3...@crash.videotron.ab.ca>
Newsgroups: rec.games.backgammon
[More Headers]
[Subscribe to rec.games.backgammon]
bj...@Lehigh.EDU wrote in message <6namum$n...@ns4-1.CC.Lehigh.EDU>...
>JG wrote:
>
>
>Last night, I opened 6/5, 24/21 off of 3-1.
Interesting play. I would play 6/2. It is important to start crowding your
opponents pieces on the ace point so they feel uncomfortable. :)
> JF rolled 4-4. I danced, JF made
>the five point, and I danced again. JF then rolled 5-5, pointed the 1, and
>went on to win a backgammon.
>
>After my opening roll, I never got to move!!!! And every time the opening
>rolls go 3-1, 4-4, I can't ever win!!!!! JF cheat$ !
To beat JF, you have to be willing to cheat yourself. After JF has taken
off a few of his pieces, save the game and edit the match file. Put all of
his extra pieces back on the 13 point. That'll teach him for cheating. :)
>
>Brad Mampe
>bj...@lehigh.edu
How instructing and enlightening is the last paragraph?? That's exactly the
kind of comments the Internet backgammon community needs...
As for the Kit Woolsey comparison, are your posts as good as his?? Except
for your "Early 33" I don't think so... You just can't compare anyone here
to him.
RODRIGO
Gary Wong
> >And I've not yet seen
> >a single world-class player complain that JF cheats. In fact,
> >every strong player I've heard speak about the subject has
> >defended the program's honesty.
>
> We never will. These guys have a lot of credibility and they could take a
> company out of business if they started saying that a program cheats. What
> would happen to Nike if Michael Jordan said their shoes are uncomfortable,
> and just plain suck???
I doubt that's much of an influence. Strong players do not
necessarily mince their words -- I've heard strong players criticise
books (eg. Becker's _Backgammon for Blood_) and software (eg. _Windows
BG_). It is instructive to note that these strong players do _not_
claim that the books/programs "just plain suck"; they tend to describe
the book/program subjectively and point out areas they believe are
weaknesses. I believe the rest of us would do very well to try to
emulate the same approach.
It is practically impossible to prove that Jellyfish does not cheat at
all. But we already have plenty of evidence to suggest that the
difference in its ability between playing with its own dice and with
manual dice (if indeed there is a difference) is very small -- too
small to be detected without a powerful statistical test. Playing 200
1-point matches with computer-generated and manual dice and inspecting
the results is a relatively powerful statistical test. Posting an
article to this group reading "The difference in playing against JF
letting it roll the dice vs. rolling one's own dice is as obvious as
day and night" is not.
There are some records of reasonably statistically significant
experiments -- somebody (I can't find the reference sorry, but I
believe it was in response to an experiment proposed by Kees van den
Doel in article <68okin$k...@cascade.cs.ubc.ca> in the thread
"Jellyfish" around the new year) played an equal number of matches
against Jellyfish with and without manual dice and posted the logs to
the newsgroup. I think the total number of games played came to
around 50 points, and the result in this case was that Jellyfish
performed marginally worse with manual dice. The conclusion was that
(for some significance level -- 95%, probably) no evidence was found
to support the hypothesis that Jellyfish's ability differed depending
on the type of dice generation in use.
Around the same time, Kit Woolsey also noted in article
<kwoolseyE...@netcom.com> that:
Well, last summer Mike Senkiewicz and Nack Ballard (who would certainly
rate as two of the best players in the world by anybody's standards) each
played Jellyfish 300 games for some pretty high stakes. The dice were
rolled manually for both sides. The final result was that Jellyfish
broke even (+58 vs. Senk, - 58 vs. Nack). This appears to be about as
good a test as one could want, and does demonstrate that with manually
rolled dice Jellyfish was able to hold its own against the best.
And yet _nobody_ has provided significant reputable data suggesting
any contrary conclusion. This ought to be a persuasive indication
that in all likelihood, Jellyfish does not cheat at all. If you feel
happier rolling manual dice, then by all means play that way! But
unless you have sufficient data from a fair experiment providing
evidence Jellyfish may cheat, it does not seem fair to ignore every
other test in which no such evidence has been found.
Murat Kalinyaprak
> Murat Kalinyaprak wrote:
>>Just in case it was misunderstood, I wasn't proposing
>>that JF gets itself out of "tough positions", etc. by
>>rolling what it needs. JF's seed/counter scheme takes
>>care of any such claims already. What I was proposing
>>was that, it "leads" the game (and also the opponent)
>>to that position.
> <----------------- perfectly blocked ---------------->
> Murat does it again! It must be deliberate, surely?
In your newsreader program, click on preferences, set
seed=5555/counter=222 and start typing. What you type
will automatically self-justify... :)
But seriously, yes it's deliberate. It's kind of like
playing cross-word puzzles, word-games, etc. It slows
down my typing a bit but also gives me another chance
and more time to re-think what I'm saying...
MK
Murat Kalinyaprak
>James Eibisch wrote:
>>Murat Kalinyaprak wrote:
>>>was that, it "leads" the game (and also the opponent)
>>>to that position.
> JE> <----------------- perfectly blocked ---------------->
> JE> Murat does it again! It must be deliberate, surely?
> Quite an achievement, I'd say.
> It's not so easy to do, as I'm
> finding out myself at this ver
> y moment! Maybe his articles a
> re computer-generated? Or mayb
> e it is a subtle indication of
> the troll-value of his posts??
> (:--------------------------:)
I don't know about their "troll-value", but my articles
are not computer generated. Although I hadn't posted in
this newsgroup until very recently, I have been posting
in other newsgroups for many years. There must be a few
thousands (literally) of my articles stored in archives
like Dejanews. I don't recall being accused of trolling
too often, if at all, and I probably have a pretty good
credibility in newsgroups where I had been posting on a
regular basis.
MK
Murat Kalinyaprak
> I'm just guessing but it sounds like you cranked the cube up to
> a high level and got lucky 3 times. Would be curious to see
> long-term results with this method.
Why do you assume it's me who ran up the cube? In fact,
it was the other way around. I had indicated this in my
first posting where I mentioned those 3 games (although
that articles didn't seem to get distributed well, some
readers may have seen it on their servers).
Nevertheless, I'm quite pleased to hear about "my luck"
vs. "JF's luck" for a change... :)
Worse news may be that I also started getting some very
strong wins against JF even when it's rolling the dice.
Yesterday I played again 3 matches with the dice rolled
by JF. The score was:
MK JF
20 0 (from 4-0)
8 16 (from 8-8)???
29 0 (from 13-0)
I'll attach the "mat" files for the 1st and 3rd matches
for your information. Unfortunately somehow I failed to
save the very match I lost. I wouldn't blame you if you
thaught that I didn't on purpose, but I'm still getting
used to saving/reloading/replaying games/matches. Also,
I kept forgetting to keep track of the seed and counter
at the beginning of each match. However, I did remember
to note them at the end of the first match, which were:
seed=13745 counter=748
One could figure out the starting counter value for the
first match by counting back the number of moves (which
I didn't bother to do). They were all consecutive games
(without modifying seed/counter) but it would be almost
impossible to recognize the dice patterns and determine
what value the last game had started on.
Also, from what I remember, I lost the last game of the
second match by only 3 pips. So, I was not "running up"
the cube without reason and 8-16 could have easily been
16-8 for me. Now, if that had happened, that would have
been some "luck"...
> And as another poster wrote. If you are that good at beating J/F
> then you should find someone to back it for money. I would be
> more that happy to back J/F at $10/point. Perhaps playing through
> FIBS and me entering the rolls to J/F at level 7-100. Let me know
> if you are interested.
Sorry, but I'm not interested in gambling beyond buying
an occasional lottery/raffle ticket, etc. I have played
backgammon since I was a kid, but not for anything more
than a round of teas, cookies and such. However, if my
apparent "luck" persists for a few dozen matches, I may
just change my mind...:)
MK
--------------------------------------------------------------
15 point match
Game 1
JellyFish : 0 Player2 : 0
1) 53: 8/3 6/3
2) 64: 8/2 6/2 51: 13/8 6/5
3) 61: 13/7 8/7 55: 13/3 13/3
4) 42: 24/20* 13/11 52: 25/20 3/1*
5) 66: Doubles => 2
6) Takes 32: 20/17* 3/1
7) 53: 25/20 11: 17/14* 14/13
8) 21: 25/23 52: 24/17
9) 63: 64: 8/2* 6/2
10) 53: 25/20 61: 17/10
11) 63: 65: 10/4 13/8
12) 63: 33: 24/21 13/10 13/10 4/1
13) 33: 51: 21/15
14) 63: 52: 15/10 8/6
15) 22: 53: 10/2
16) 53: 25/20 13/10 63: 10/4 10/7
17) 21: 20/18* 13/12 31: 25/21
18) 54: 18/13 10/6 52: 8/3 8/6
19) 62: 20/12 51: 6/1 2/1
20) 44: 12/4* 12/4 65: 25/14
21) 53: 13/8 13/10 53: 14/6
22) 44: 20/12 7/3 7/3 66: 6/0 6/0 6/0 6/0
23) 53: 20/12 43: 4/0 3/0
24) 11: 12/11 12/11 8/6 61: 3/0 1/0
25) 62: 11/5 11/9 22: 3/0 2/0 2/0
26) 53: 10/5 9/6 54: 1/0 1/0
27) 62: 6/0 2/0 43: 1/0 1/0
Wins 2 points
Game 2
JellyFish : 0 Player2 : 2
1) 42: 8/4 6/4
2) 62: 24/18 13/11 32: 6/1*
3) 61: 25/24* 24/18 54: 25/20 24/20
4) 64: 13/7 11/7 31: 24/20
5) 62: 13/7 6/4 33: 13/10 13/10 8/5 8/5
6) 42: 7/3 6/4 63: 20/11
7) 65: 13/7 13/8 32: 6/3 13/11
8) 62: 8/2 8/6 21: 6/3
9) 54: 8/3 8/4 61: 11/5 11/10
10) 63: 7/1 4/1 21: 6/3
11) 53: 7/2 7/4 53: 10/2
12) 64: 18/8 61: 13/7* 3/2
13) 44: Doubles => 2
14) Takes 32: 20/17* 10/8
15) 55: 21: 8/6 7/6
16) 65: 55: 20/15 17/7 13/8
17) 54: 61: 10/4 8/7
18) 55: 21: 15/12
19) 31: 25/24 31: 5/1*
20) 64: 11: 7/6 4/1
21) 51: 33: 12/6 7/1
22) 65: 42: 6/2 6/4
23) 43: 61: 6/0 6/5
24) 51: 21: 2/0 1/0
25) 65: 25/19 53: 5/0 4/1
26) 62: 25/17 52: 5/0 2/0
27) 66: 19/7 17/5 32: 3/0 2/0
28) 54: 7/2 4/0 44: 5/1 4/0 4/0 3/0
29) 33: 6/0 3/0 3/0 61: 1/0 1/0
30) 64: 6/0 4/0 44: 1/0 1/0
Wins 2 points
Game 3
JellyFish : 0 Player2 : 4
1) 61: 13/7 8/7 31: 8/5 6/5
2) 41: 13/9 24/23 63: 24/15
3) 43: 13/9 13/10* 43: 25/21 24/21
4) 32: 13/10 23/21 42: 8/4* 6/4
5) 65: 62: 13/5
6) 51: 25/24 6/1 54: 13/8 13/9
7) 65: 24/13 53: 9/1*
8) 63: 25/22 13/7 32: 6/3* 5/3
9) 32: 25/23 6/3 55: 13/3 13/3
10) 54: 23/14 51: 8/2
11) 31: 7/3 64: 8/2 5/1
12) 63: 14/5 64: 5/1
13) 42: 7/3 7/5 32: 6/3 6/4
14) 41: 6/2 3/2 32: 4/1 3/1
15) Doubles => 2 Takes
16) 65: 8/2 8/3 Doubles => 4
17) Takes 64: 21/11
18) Doubles => 8 Takes
19) 31: 5/4* 4/1 Doubles => 16
20) Takes 52: 25/20* 20/18
21) 21: 43: 18/11
22) 41: 22: 11/7 11/7
23) 31: 52: 7/2 3/1
24) 66: 25/7 10/4 43: 7/0
25) 61: 10/4 6/5 64: 4/0 4/0
26) 31: 9/6 7/6 42: 3/0 2/0
27) 65: 9/4 6/0 21: 2/0 1/0
28) 21: 2/0 1/0 63: 3/0 3/0
29) 61: 6/0 1/0 64: 2/0 1/0
30) 31: 3/0 2/1 21: 1/0 1/0
31) 21: 2/0 1/0 44: 1/0 1/0
Wins 16 points and the match
--------------------------------------------------------------
15 point match
Game 1
JellyFish : 0 Player2 : 0
1) 54: 13/8 24/20 65: 24/13
2) 33: 8/5 8/5 6/3 6/3 51: 13/8 24/23
3) 42: 24/20 13/11 22: 13/11 13/11 6/4 6/4
4) 22: 20/16 20/16 54: 23/14*
5) 44: Doubles => 2
6) Takes 42: 14/10 13/11
7) 51: 25/20 16/15* 21: 25/23 11/10*
8) 32: 25/20 42: 13/9* 13/11
9) 62: 25/23 8/2* 44: 25/17* 17/9
10) 52: 25/23 20/15* 21: 25/23* 11/10*
11) 64: 42: 23/17
12) 62: 25/23 51: 10/5* 6/5
13) 64: 21: 9/7 8/7
14) 63: 25/22 21: 11/9 8/7
15) 66: 51: 11/5
16) 31: 25/22 23/22 33: 17/8 7/4
17) 43: 13/9 13/10 21: 9/7 9/8
18) 62: 22/16 6/4 31: 8/5 7/6
19) 21: 16/14 9/8 33: 8/5 8/5 8/5 4/1
20) 41: 8/4 13/12 66: 7/1 7/1 6/0 6/0
21) 61: 14/7 42: 6/0
22) 61: 22/16 10/9 41: 5/1 5/4
23) 51: 12/7 9/8 63: 5/0 4/1
24) 42: 8/2 63: 5/0 4/1
25) 43: 16/9 54: 5/0 4/0
26) 31: 9/5 66: 5/0 5/0 1/0 1/0
27) 52: 23/18 22/20 42: 1/0 1/0
28) 21: 20/18 7/6 31: 1/0 1/0
Wins 6 points
Game 2
JellyFish : 0 Player2 : 6
1) 42: 8/4 6/4
2) 53: 8/3 6/3 42: 24/20 13/11
3) 31: 8/5* 6/5 41: 25/21 24/23
4) 53: 13/8 13/10 55: 23/13 21/11
5) 43: 24/20 13/10 22: 13/9 13/9
6) 65: 20/9 54: 13/4
7) 64: 10/4 8/4 66: 13/7 13/7 11/5 11/5
8) 43: 9/5 10/7 Doubles => 2
9) Drops Wins 1 point
Game 3
JellyFish : 0 Player2 : 7
1) 64: 8/2 6/2
2) 51: 13/8 6/5 61: 13/7 8/7
3) 54: 24/15 32: 13/10* 10/8
4) 66: Doubles => 2
5) Takes 42: 24/20* 24/22
6) 43: 25/21 25/22 52: 20/13
7) 62: 13/7 24/22 43: 22/18* 18/15
8) 54: 25/21 8/3 41: 13/8
9) 32: 13/10* 6/4 54: 25/21* 21/16
10) 65: 25/14 52: 16/11* 13/11
11) 21: 25/22 62: 13/7 13/11
12) 31: 8/5 6/5 63: 11/5 8/5
13) 31: 13/10 13/12 63: 11/5 11/8
14) 61: 22/16 12/11 51: 6/1 6/5
15) 32: 10/7 11/9 53: 7/2 8/5
16) 53: 8/3 7/4 52: 7/2 7/5
17) 64: 10/4 16/12 44: 5/1 5/1 5/1 5/1
18) 65: 8/2 12/7 32: 8/5 8/6
19) 31: 9/6 3/2 52: 5/0 2/0
20) 62: 21/13 21: 2/0 1/0
21) 61: 22/15 11: 1/0 1/0 1/0 1/0
22) 61: 15/9 13/12 64: 6/0 6/2
23) 64: 9/3 12/8 22: 6/4* 4/0 2/0
24) 51: 25/24 6/1 51: 5/0 5/4
25) 52: 22/15 61: 4/0
26) 21: 15/13 8/7 22: 2/0 2/0
Wins 6 points
Game 4
JellyFish : 0 Player2 : 13
1) 62: 24/16
2) 21: 24/21 51: 16/11 24/23
3) 65: 21/10 53: 23/15*
4) 42: 25/21 6/4 43: 8/4* 4/1*
5) 55: 25/10* 25/20 32: 25/22 13/11
6) 64: 10/4 20/16 51: 13/8 6/5
7) 21: 6/3* 66:
8) Doubles => 2 Takes
9) 21: 6/3 11: 25/24 11/9* 6/5
10) 22: 25/21 3/1* 3/1 43: 25/22 13/9
11) 63: 21/15 13/10 32: 13/10* 13/11
12) 62: 25/23 10/4 Doubles => 4
13) Takes 63: 22/13
14) Doubles => 8 Takes
15) 52: 23/18 8/6 Doubles => 16
16) Takes 61: 13/7* 8/7
17) 44: 25/21 13/5 13/9 22: 10/4* 6/4
18) 61: 25/24* 13/7 11:
19) 61: 13/7 6/5 62: 25/23 11/5
20) 64: 9/3 7/3 54: 23/18* 18/14
21) 62: 25/23 8/2 53: 11/6 5/2*
22) 43: 25/22 6/2 53: 6/1* 14/11
23) 51: 25/24* 6/1 21:
24) 51: 8/2 11:
25) 61: 2/1 42:
26) 41: 5/1 5/4 52: 25/20 11/9
27) 33: 4/1 4/1 4/1 4/1 65: 9/3* 20/15
28) 63: 25/22* 61: 25/18
29) 41: 24/23* 53: 25/20 15/12
30) 22: 3/1 3/1 64: 12/6 18/14
31) 42: 44: 9/5 7/3* 7/3 6/2*
32) 31: 25/24 63: 8/2 5/2
33) 64: 66: 20/2 9/3
34) 64: 54: 14/9 8/4
35) 51: 25/24 61: 9/2
36) 21: 2/1 64: 6/0 6/2
37) 52: 24/17 43: 4/0 3/0
38) 21: 17/15 2/1 41: 4/0 4/3
39) 53: 15/7 52: 5/0 5/3
40) 43: 24/20 7/4 41: 3/0 2/1
41) 31: 20/16 53: 3/0 3/0
42) 55: 16/1 4/0 42: 3/0 2/0
43) 63: 1/0 1/0 65: 2/0 2/0
44) 51: 1/0 1/0 63: 2/0 2/0
45) 66: 1/0 1/0 1/0 1/0 53: 1/0
Wins 16 points and the match
--------------------------------------------------------------
Murat Kalinyaprak
> James, I think you could be on to something here.
> The chances of something like this occurring in a
> above. Still, strange things happen. For instance
> I saw recently reported in the English press that
> a woman in West London had sliced a tomato in two
> only to discover that it had an important passage
> from the Koran visible in the pattern of the pips.
Well done justifying the text, Rew. But what got into
you to add this garbage at the end...?
As my name might have provided you a hint, I was born
and raised in Turkey as a Muslim. I'm not a religious
person and I don't feel a need to mock at such things
like bleeding/weeping Jesus statues, I wonder whether
I shouldn't still ask if the tomato you mentioned was
imported from Mecca. If not, for someone who believes
in seeds and counters, wouldn't you rather think that
the odds would be much higher for some English tomato
seed to contain Biblical revelations from Jesus...?
MK
Patti Beadles
Rodrigo Andrade <candrade@_R_E_M_O_V_E_wt.net> wrote:
>We never will. These guys have a lot of credibility and they could take a
>company out of business if they started saying that a program cheats.
Oh, puhleez. I know a fair number of world-class backgammon players,
and I can't think of a single one of them who would hesitate to put
forth a legitimate criticism of Jellyfish, or any other backgammon
merchandise.
They wouldn't, however, shoot from the hip about it. Which is what
most people here are doing.
-Patti
--
Patti Beadles | Not just your average
pat...@netcom.com/pat...@gammon.com | degenerate gambling adrenaline
http://www.gammon.com/ | junkie software geek leatherbyke
or just yell, "Hey, Patti!" | nethead biker.
Rew Francis
Murat, it was not my intention to cause offence to
you or anyone else with this post - can I offer my
apologies if any was taken. Your name did not have
any bearing on my choice of this example - I chose
it simply because it was another case of extremely
unlikely coincidence, which happened to come to my
mind at the time. As for the origin of the tomato,
I am not able to comment - I don't recall it being
mentioned in the article. However, from looking at
the packaging on tomatoes I have bought, I believe
most that you can buy in England are grown locally
or else imported from Spain or the Canary Islands.
Rew
Robert-Jan Veldhuizen
MK> Robert-Jan Veldhuizen wrote:
MK>
>>James Eibisch wrote:
MK>
>>>Murat Kalinyaprak wrote:
MK>
>>>>.....................................................
>>>>was that, it "leads" the game (and also the opponent)
>>>>to that position.
>> JE> <----------------- perfectly blocked ---------------->
>> JE> Murat does it again! It must be deliberate, surely?
MK>
>> Quite an achievement, I'd say.
>> It's not so easy to do, as I'm
>> finding out myself at this ver
>> y moment! Maybe his articles a
>> re computer-generated? Or mayb
>> e it is a subtle indication of
>> the troll-value of his posts??
>> (:--------------------------:)
MK> I don't know about their "troll-value", but my articles
MK> are not computer generated. Although I hadn't posted in
MK> this newsgroup until very recently, I have been posting
MK> in other newsgroups for many years. There must be a few
MK> thousands (literally) of my articles stored in archives
MK> like Dejanews. I don't recall being accused of trolling
MK> too often, if at all, and I probably have a pretty good
MK> credibility in newsgroups where I had been posting on a
MK> regular basis.
Well it wasn't an accusation...I like trolls. :)
.....<----------------------------------------------------->
But again, what's the secret?
--
Zorba/Robert-Jan
Rodrigo Andrade
It is. Well, maybe not for famous backgammon players. But rich and famous
people always have influence over mortals like us.
>they tend to describe
>the book/program subjectively and point out areas they believe are
>weaknesses.
Yep, but it's easier to subjectively describe a book than a Nike shoe.
RODRIGO
Rodrigo Andrade
>1-point matches with computer-generated and manual dice and inspecting
>the results is a relatively powerful statistical test
That's an interesting point I've been observing over the last few months.
Playing, say, a bunch of single-point matches till you get to 5 points is
easier than playing a 5-point match. JF cube handling in the single pointers
is poorer, simply because it doesn't know (it couldn't) you're planning to
play other matches.
RODRIGO
Gary Wong
> >Playing 200
> >1-point matches with computer-generated and manual dice and inspecting
> >the results is a relatively powerful statistical test
>
> That's an interesting point I've been observing over the last few months.
> Playing, say, a bunch of single-point matches till you get to 5 points is
> easier than playing a 5-point match.
I didn't quite mean to imply anything about any score being "easier". The
reason I suggested playing one-pointers is only because the cube adds a
HUGE amount of variance to the result. The 300 money games against
Jellyfish I alluded to in another message is a good example of this -- even
though in both cases one player came out 58 points ahead, this is not
sufficient evidence that that player is better. (I can give the maths
if anybody cares, but it's fairly boring so I won't bother with it now).
This is because there is a very good chance that those 58 points came
from a few games where the cube reached 16, 32 or higher. All of the
other hundreds of games where the cube was dropped at 1 hardly count
toward the final score -- even though they probably contained nearly
as much `skill' (in an information theoretic sense, they carry nearly
as much signal about who is the better player).
On the other hand, a lead of 58 points after 300 1-pointers (ie. playing
300 one point games, and coming out leading 179 to 121) _is_ a significant
result (because the variance of a cubeless, gammonless game is far less
than an ordinary money one). That's what I mean about a large number of
one-pointers being a powerful statistical test. Unfortunately, it's
not a powerful test of cube handling (obviously) but most people would
agree that a person's skill at one-point matches corresponds reasonably
well to their overall skill in money games or longer matches.
> JF cube handling in the single pointers
> is poorer, simply because it doesn't know (it couldn't) you're planning to
> play other matches.
I'm not sure I understand you. There's NO cube handling in single pointers
(no gammons either, for that matter) -- so how can JF's cube handling be
poorer?
Rodrigo Andrade
Nor did I. That was just an observation that had nothing to o w/ your
post... just wanted to hear people's thoughts.
>I'm not sure I understand you. There's NO cube handling in single pointers
Yes, there is. Just turn on the option Use Cube from the Settings menu. I
know it's an incoherent thing, but JF does use the cube on 1-pointers.
RODRIGO
Harold Evenson
Rodrigo Andrade wrote in message <6oo709$qrd$1...@news4.wt.net>...
>Better yet: I ran "Brad Mampe" and look what I found: 8 out of 18 matches
to
>rec.games.backgammon (the other 10 to rec.games.bridge, if anyone's
>wondering...). Also some of these posts actually belong to people who
quoted
>him. Now, let's examine some of these 8 bites of wisdom:
[snip quoted material]
>HEEEEEEEEEEEEEEEEEEEEY!!!!!!!!!!!!!!! It seems like I'm not the only one
>questioning the opinion that JF cheats, eh??? Let's move on:
[snip]
At your earliest possible convenience you should get a dictionary and look
up the words "SARCASM" and "HYPERBOLE". The comments you quoted were
written in that style.
If you are too thick to understand it, let me be clear. Brad suggested
making one of the worst possible opening moves for that roll, and went on to
suggest (sarcastically) that Jellyfish was cheating because it went on to
win. The real intended point of the message (which seems to have escaped
you thus far) is that your bad plays can make JF's good plays seem
unrealistically good.
In my followup, I think I made even more outrageous statements, just to
point out how absurd the JF cheats argument had become. It should be clear
that they weren't to be taken literally.
News1
Rodrigo Andrade wrote in message <6op76v$a8l$1...@news4.wt.net>...
ehhh... a one point match is worth one point...it doesn't matter if Jelly or
you jack up the cube to 512...the winner of the match wins 1-0... sure you
can toss a cube...it doesn't mean anything though.
Julian
"ggsysop" on GamesGrid
"jhaley" on GamesGrid
"GGguide11" on GamesGrid
Adrian Pelland
Kalinyaprak) wrote:
>In <35ACEE9C...@elender.hu> Dean Kezan wrote:
>
>>P.S. If you are really interested for JF dices try in some
>>really though sitution (exp. JF need only 4-X to come in
>>from bar and it get it 4-5)try to use second button from
>>master dice,to repeat the rolls. Some times it can be
>>really unbelievable ( 4-4 , 4-5, 4-6 etc.)
>
>Just in case it was misunderstood, I wasn't proposing
>that JF gets itself out of "tough positions", etc. by
>rolling what it needs. JF's seed/counter scheme takes
>care of any such claims already. What I was proposing
>was that, it "leads" the game (and also the opponent)
>to that position.
>
>I have a feeling that JF also relies on its opponents
>to make the moves expected from them, at the selected
>difficulty level. When JF is rolling the dice, it may
>perhaps be thrown off by a purposefully made bad move
>temporarily but the effect may not last long since it
>may/will make its following move based on to the next
>roll to come again.
>
>Speculating and experimenting with such things is fun
>and perhaps the developers of JF may be having a ball
>reading what's written on this subject in this group.
>
>What seems to be sufficiently clear to a lot of us is
>that, JF plays much better when it rolls the dice and
>not all that good when the dice is rolled manually...
>
>MK
>
It's obvios to any one who studies the game what Jelly Fish is doing.
I am using the shareware version and conclusions are easily come by,
by analysing the rolls when on the bar or after JF cubes. Check it out
for yourself. ps ( where are these top players defending JF in this
newsgroup.)
Gary Wong
> Worse news may be that I also started getting some very
> strong wins against JF even when it's rolling the dice.
> Yesterday I played again 3 matches with the dice rolled
> by JF. The score was:
>
> MK JF
> 20 0 (from 4-0)
> 8 16 (from 8-8)???
> 29 0 (from 13-0)
>
> I'll attach the "mat" files for the 1st and 3rd matches
Thanks very much for posting the games. It's good to see substantiave
data rather than vague accusations. It's interesting to see that every
match was won by the cube going high (8 or 16) and each match contained
a small number of games -- unusual for 15 point matches. I suspect the
cube only got this high because you doubled in situations I wouldn't consider
correct doubles (don't take this as criticism, it's only an opinion) and
if Jellyfish considers the same position a double on _its_ side, then the
cube will keep getting turned and the game turns into double match point:
effectively, you end up playing one-pointers.
Assuming this is the case (it might not be, but humour me :-), and let's
estimate that Jellyfish is doubling when it has 2/3 cubeless winning
chances, then your match equity becomes the chance of winning this game
(ie. 33%) instead of what you would expect if your ability is equal to
JF's (50%) -- this presumed `incorrect' cube behaviour is costing you
17% match equity.
> 15 point match
>
> Game 1
> JellyFish : 0 Player2 : 0
> 9) 63: 64: 8/2* 6/2
> 10) 53: 25/20 61: 17/10
> 11) 63: 65: 10/4 13/8
> 12) 63: 33: 24/21 13/10 13/10 4/1
> 13) 33: 51: 21/15
> 14) 63: 52: 15/10 8/6
> 15) 22: 53: 10/2
Isn't this sufficient disproof that Jellyfish consistently rolls "miracle
numbers" to enter from the bar? The odds are slightly in favour of entering
a four point board (20/36) and yet here it fails 6 times out of 7. An
isolated incident means nothing overall to be sure, but I can imagine
many humans getting very hot under the collar if JF rolled dance numbers
6 times out of 7 for them!
> Game 3
> JellyFish : 0 Player2 : 4
> 1) 61: 13/7 8/7 31: 8/5 6/5
> 2) 41: 13/9 24/23 63: 24/15
> 3) 43: 13/9 13/10* 43: 25/21 24/21
> 4) 32: 13/10 23/21 42: 8/4* 6/4
> 5) 65: 62: 13/5
> 6) 51: 25/24 6/1 54: 13/8 13/9
> 7) 65: 24/13 53: 9/1*
> 8) 63: 25/22 13/7 32: 6/3* 5/3
> 9) 32: 25/23 6/3 55: 13/3 13/3
> 10) 54: 23/14 51: 8/2
> 11) 31: 7/3 64: 8/2 5/1
> 12) 63: 14/5 64: 5/1
> 13) 42: 7/3 7/5 32: 6/3 6/4
> 14) 41: 6/2 3/2 32: 4/1 3/1
> 15) Doubles => 2 Takes
> 16) 65: 8/2 8/3 Doubles => 4
Is this really a redouble?
+13-14-15-16-17-18-------19-20-21-22-23-24-+
| X X | | X X O X X X |
| X X | | X X O X X |
| | | X X |
| | | |
| | | |
v| |BAR| | O on roll, 2 cube.
| | | O | X: 15 away
| | | O O | O: 13 away
| | | O O |
| | | O O O O | Pipcount: X 76, O 71
| | | O O O O |
+12-11-10--9--8--7--------6--5--4--3--2--1-+
I don't see how it can be a redouble without enough market losers (really
only 44). O has plenty of bad rolls (16 combinations of 1, 2, 5 and 6)
and many more forcing him to run with a single chequer leaving at least
a double shot. The race is close (O is a little ahead and has the
advantage of being on roll, but is 2 crossovers behind). But O has
essentially no chance of winning a gammon here even if hits a shot
(crunched board) where it's easy to imagine many sequences to X winning
a gammon. If you ask me, this all adds up to "no redouble!".
> --------------------------------------------------------------
> 15 point match
>
> Game 4
> JellyFish : 0 Player2 : 13
> 1) 62: 24/16
> 2) 21: 24/21 51: 16/11 24/23
> 3) 65: 21/10 53: 23/15*
> 4) 42: 25/21 6/4 43: 8/4* 4/1*
> 5) 55: 25/10* 25/20 32: 25/22 13/11
> 6) 64: 10/4 20/16 51: 13/8 6/5
> 7) 21: 6/3* 66:
> 8) Doubles => 2 Takes
> 9) 21: 6/3 11: 25/24 11/9* 6/5
> 10) 22: 25/21 3/1* 3/1 43: 25/22 13/9
> 11) 63: 21/15 13/10 32: 13/10* 13/11
> 12) 62: 25/23 10/4 Doubles => 4
Why on earth are you doubling here? It's a 15 point match, and you're
leading 13 to 0 -- you don't WANT the cube to go above 2!
In any case, congratulations on your wins -- winning 2 out of 3 15 point
matches against Jellyfish, including 2 backgammons, is quite an impressive
achievement :-)
my_...@hotmail.com
Murat Kalinyaprak <mu...@cyberport.net> wrote:
>
> Why do you assume it's me who ran up the cube? In fact,
> it was the other way around.
>
>
> Game 3
> JellyFish : 0 Player2 : 4
> 1) 61: 13/7 8/7 31: 8/5 6/5
> 2) 41: 13/9 24/23 63: 24/15
> 3) 43: 13/9 13/10* 43: 25/21 24/21
> 4) 32: 13/10 23/21 42: 8/4* 6/4
> 5) 65: 62: 13/5
> 6) 51: 25/24 6/1 54: 13/8 13/9
> 7) 65: 24/13 53: 9/1*
> 8) 63: 25/22 13/7 32: 6/3* 5/3
> 9) 32: 25/23 6/3 55: 13/3 13/3
> 10) 54: 23/14 51: 8/2
> 11) 31: 7/3 64: 8/2 5/1
> 12) 63: 14/5 64: 5/1
> 13) 42: 7/3 7/5 32: 6/3 6/4
> 14) 41: 6/2 3/2 32: 4/1 3/1
> 15) Doubles => 2 Takes
> 16) 65: 8/2 8/3 Doubles => 4
> 17) Takes 64: 21/11
> 18) Doubles => 8 Takes
> 19) 31: 5/4* 4/1 Doubles => 16
> 20) Takes 52: 25/20* 20/18
[...]
Now, I'm no expert, but I believe you're better off
waiting until you're actually favoured to win the game
before doubling.
>
> Game 4
> JellyFish : 0 Player2 : 13
> 1) 62: 24/16
> 2) 21: 24/21 51: 16/11 24/23
> 3) 65: 21/10 53: 23/15*
> 4) 42: 25/21 6/4 43: 8/4* 4/1*
> 5) 55: 25/10* 25/20 32: 25/22 13/11
> 6) 64: 10/4 20/16 51: 13/8 6/5
> 7) 21: 6/3* 66:
> 8) Doubles => 2 Takes
> 9) 21: 6/3 11: 25/24 11/9* 6/5
> 10) 22: 25/21 3/1* 3/1 43: 25/22 13/9
> 11) 63: 21/15 13/10 32: 13/10* 13/11
> 12) 62: 25/23 10/4 Doubles => 4
> 13) Takes 63: 22/13
> 14) Doubles => 8 Takes
> 15) 52: 23/18 8/6 Doubles => 16
> 16) Takes 61: 13/7* 8/7
One thing you should be made aware of is that Jellyfish,
like most backgammon players, does not care if it loses
a 15 point match 15-0 or 29-0. Therefore, when you're
already ahead 13-0, it's generally not the correct
strategy to raise the cube from 2 to 4, and again from
8 to 16.
Your original comment, that Jellyfish tends to roll
exactly what it needs, was disproven in spectacular
fashion based on these two games.
my_...@hotmail.com
apel...@netra.mysterynet.mb.ca (Adrian Pelland) wrote:
>
> It's obvios to any one who studies the game what Jelly Fish is doing.
> I am using the shareware version and conclusions are easily come by,
> by analysing the rolls when on the bar or after JF cubes. Check it out
> for yourself. ps ( where are these top players defending JF in this
> newsgroup.)
>
Stock Reply:
You can use the seed/counter option to verify that the rolls of the
dice are *independent* of the current position of the board. Check
it out for yourself.
As for your second question, I imagine most players are tired of this
discussion on here. I rarely read this newsgroup, but when I
do, it's always the same old story...
Gary Wong
> ps ( where are these top players defending JF in this newsgroup.)
They're sick of it and have got better things to do. There have been many
excellent defences posted in the past. Read question 10 in the mini-FAQ
at http://www.cityraccoon.com/minifaq.html for a list of some of them.
Dean Kezan
You little hurt my feelings with your previous post about "Proof JF
cheat".
I had feeling just because I made post about JF dice generator that mean I
am not
honest or that, I am "lucky player".
But than I remember one article where two champion's ( excellent player's)
played with JF. It was bolded that they rolled manually dice's.
What this proof ?
That players was superstition on computer dice's or it is nicer to rolled
manually.
As I remember result was: one player won against JF and second loosed.
This is also a proof that JF don't cheat on strategy or playing, which I
never doubt.
Am I right until now ?
Take an example.
If one of best human player would play against middle player and on some
circumstances
middle player win ( hard but mathematically possible ) all watchers would
say that middle player is lucky or "lucky player".
Right ?
So take this fact and put it our case "JF against human".
That mean that all humans are "lucky player" against JF.
So get back on the point.
In this post there was lot of complainant and defenses also word "cheat".
Is this fact , proof anything ?
It proof that there is suspicion on dice's which sometimes can be very
frustrated.
Frustrated in this point. If you have 50 % chances to come in from bar and
you just can't do it in 3 rolls and JF have only 1 possibility and come in
at first roll.
I can also prove that Mr. Author of JF is not guilty on that fact.
(from previous posts, just matter of puzzle)
But can someone prove even Mr. Author, that dice generator don't have
anomalies in calculation ?
Anomalies can be ( mathematically can be proofed) in poor dice formula so
for an example: small numbers of yours can be big numbers of opposite ,or
if you have big numbers it is also possible that you will get big numbers
,or if you ... until the formula is "balanced" ( depend on type of
formula).
It wouldn't be first time in mankind that such failures are not noted at
first place especially in computer technology because everything are
numbers and human brain have difficulty to control it.
Proof Intel with Pentium processor (first edition), Microsoft with first
release of 95 ( lot of bugs)
I also claim that backgammon is game of skill with factor of luck (the
reason why i like it)
and it is better to play with better player or with player of same level (
more than one reason).
When I lose against human I never been angry because I had good advisory
and I learned something. Also I have pleasure ( fillness) in myself but in
JF I don't have even pleasure ( emptiness).
I think there are lot of people out there who have the same feeling about
the JF.
And I would like that people like you , Mr.Goodwin or Mr.Author pay
attention on our complains because something is wrong and it would be the
best if we could solve this problem as sooner is possible and painless.
Thank you for your time and take minute to think about it.
Well I hope that I will have opportunity to test my skill's against yours.
Best regard's Dean ke...@elender.hu
Murat Kalinyaprak
>apel...@netra.mysterynet.mb.ca (Adrian Pelland) writes:
>> ps ( where are these top players defending JF in this newsgroup.)
>They're sick of it and have got better things to do.
I wouldn't have any difficulty believing this. For
the past several days, I have spent so much time
on this JF stuff that I'm starting to ask myself
whether I shouldn't have better things to do also...
>There have been many excellent defences posted in the past.
>Read question 10 in the mini-FAQ at
>http://www.cityraccoon.com/minifaq.html for a list of some of them.
Following this link, all I could fine was 4 or 5
articles specificly about JF. In all those, only
the argument that JF plays a given board always
the same way deserves any merit. However, that
argument goes against JF, in that it effectively
defeats any claims about JF's using neural/fuzzy/etc.
logic. Since what one observes doesn't quite jive
with what would be considered "random" by some
people, one keeps going back to questioning this
seed/counter scheme and whether there may be any
correlation between it and board positions, etc.
For example, JF allows the seed to be "999999999"
and the counter to be "99999". That would give
"99,999,999,999,999" conbinations. We know that by
manually setting the seed/counter, one can replicate
a certain "sequence" of dice rolls. Let's say that
an average game lasts 60-70 or even 100 rolls. You
certainly would want to avoid repeating the same
sequences and giving the player the impression of
"deja-vue" too often. But what's the purpose of
such large numbers? Are we supposed to somehow
believe that JF can generate 999,999,999,999
unique sequences of 100 dice rolls...? (which would
require either a hell-of-an algorithm or crunching
power of a "Cray" supercomputer). If there is no
intention to eliminate repeating sequences, what's
the big numbers good for? (other than perhaps an
attempt to "make-believe"?). Personally, I think
the number of sequences are finite, perhaps quite
small and circular, with thousands of seed/counter
combinations mapping to the same entry points. If
this can be determined, then it may be worthwhile
to look further into whether a finite number of
board positions somehow correlate to a finite
number of seed/counter combinations. While I admit
that it sure would take a genius to accomplish
such a thing, I'm not in a hurry to rule it out.
I'm typing these as fast as the thoughts go through
my mind and my reasoning may well be flawed. But I
would like to hear if anybody has further comments
in this area/direction...?
MK
Murat Kalinyaprak
>In <6ojm06$nbr$1...@news.chatlink.com> Murat Kalinyaprak wrote:
>>I have read articles where some writers were arguing
>>that JF's seeming luck is due to its positioning its
>>pieces in such a way to make better usage out of the
>>next roll (no matter what it rolls). Of course, this
>>is a naive argument in that it assumes the other guy
>>is picking berries in the hills, while JF is placing
>>its pieces in optimum spots as it wishes. It totally
>>ignores the fact that, the other player is trying to
>>do the very same thing also.
>And the fact is, Jellyfish is *much* better at it than
>anybody who's been complaining here.
And may I ask how do you know this "fact"...?
Take me as an example of the ones who have been
"complaining here". Have you ever seen me play?
Do you know how good/bad I may be at positioning
my pieces? The answer is "no". Yet, you present
such an argument as a "fact"...?
>Jellyfish is better than (pulling a number out of the
>air) 99% of the backgammon players in the world.
As I'm coming to find out, so far I seem to be
doing as good or even better than JF, whether with
automatic or manual dice rolls. I wish that this
could make me 99% or 99.000001% better than all
of the players in the world. But somehow, I just
can't bring myself to believe such a thing. The
little feeling of accomplishment that I can draw
from it, is that I may be as good or better than
"just another computer backgammon game"... :(
I think that elevating/glorifying JF to such an
extent is no more than grossly exaggerating. If
the above "99%" was a "fact" instead of being a
"number you pulled out of the air", I might have
considered changing careers...
MK
Patti Beadles
Murat Kalinyaprak <mu...@cyberport.net> wrote:
>Take me as an example of the ones who have been
>"complaining here". Have you ever seen me play?
>Do you know how good/bad I may be at positioning
>my pieces? The answer is "no". Yet, you present
>such an argument as a "fact"...?
There's a story about a grand master chess player who is on
a train and comes across a man with a chess board in front of
him. He sits down and asks the man if he'd like to play a
game.
The man agrees, and asks what stakes the master would like to play
for.
"Any amount you choose", replies the master.
"WHAT?!? Do you have any idea who I am?"
"No, none at all. And that's why I'll let you choose the stakes."
-Patti
--
Patti Beadles |
pat...@netcom.com/pat...@gammon.com |
http://www.gammon.com/ | "I trust you. It's just
or just yell, "Hey, Patti!" | that I'm scared of you."
Patti Beadles
number generation. The questions you ask are all relatively basic
within that field.
Yes, you're partially right. Generally, the way RNGs work is that
there's some magic algorithm that's carefully crafted to produce a
(usually) 32-bit random number. The algorithm is initialized with
some number, called a seed. After the first random number is
produced, that number is then used as part of the equation to get the
next number. And so on, ad inifinitum. Eventually the algorithm
will wrap around on itself, but for a 32-bit random number that's not
going to happen very quickly.
So yes, there's a circular sequence. BUt it will be millions if not
billions of numbers long before it wraps. Seed X counter basically
means initialize the random number generator with X, and then start
at the Yth roll after that.
That doesn't necessarily mean that Jellyfish can generate that huge
number of unique sequences of 100 rolls. It's almost certain that
there are sequences of numbers in there that will repeat 50-100 rolls
before diverging again.
So yes, you're partially right. It is just a single stream of
numbers, but it's tremendously large and circular, not small and
circular.
Rodrigo Andrade
** newsreader won't allow me to post that much quoted stuff **
There you go... I've posted something like that, but it seems that my
server didn't post it. Even though it's possible to have 99,999,999,999,999
different combinations of seeds and counters, the possible number of
combinations between 2 6-faced dice, and the sequences of rolling 100, or
200, or 300 rolls are limited!!!! It's impossible to see a game in which
rolls never repeat. True, the seed and counter didn't have to use such long
fieldwidth.
RODRIGO
Gary Wong
> In <wtbtqna...@brigantine.CS.Arizona.EDU> Gary Wong wrote:
> >apel...@netra.mysterynet.mb.ca (Adrian Pelland) writes:
> >> ps ( where are these top players defending JF in this newsgroup.)
> >They're sick of it and have got better things to do.
>
> I wouldn't have any difficulty believing this. For
> the past several days, I have spent so much time
> on this JF stuff that I'm starting to ask myself
> whether I shouldn't have better things to do also...
I guess the fact that I'm still bothering to follow-up suggests that I
don't have better things to do either. Maybe that's also why I'm not
a top player :-)
> >There have been many excellent defences posted in the past.
> >Read question 10 in the mini-FAQ at
> >http://www.cityraccoon.com/minifaq.html for a list of some of them.
>
> Following this link, all I could fine was 4 or 5
> articles specificly about JF. In all those, only
> the argument that JF plays a given board always
> the same way deserves any merit. However, that
> argument goes against JF, in that it effectively
> defeats any claims about JF's using neural/fuzzy/etc.
> logic.
That's not true at all. It uses a neural net evaluation function but not
fuzzy logic. Playing the same way for a given position is exactly what
you would expect from a neural net (given that it has already been fully
trained and so its weights never change) -- the neural net is a deterministic
function whose outputs depends only on its inputs; and the inputs are
entirely derived from the board state. You might want to read the references
on neural nets and other evaluators as applied to backgammon at:
http://forum.swarthmore.edu/~jay/learn-game/systems/gammon.html
Most of the references there concern TD-Gammon which was the first
genuinely `expert' neural net player. There are several publications
in highly repected journals like the Communications of the ACM, and
in-depth analyses of its playing strength including many comments by
and scores of matches against top human players. It is almost
inconceivable that TD-Gammon could "cheat" and stand up to this level
of scrutiny. Jellyfish and TD-Gammon use a very similar approach and
are of comparable strength, so if you can believe that TD-Gammon can
play this well without "cheating", then why not Jellyfish? Snowie and
loner (other neural nets) and the latest Motif (not a NN) also play at
various expert levels (within about 100 rating points of the best
humans) -- do these programs all cheat too?
> Since what one observes doesn't quite jive
> with what would be considered "random" by some
> people, one keeps going back to questioning this
> seed/counter scheme and whether there may be any
> correlation between it and board positions, etc.
How can there be? You can set the seed and counter manually, and you can
change the board position too. They are utterly independent.
> For example, JF allows the seed to be "999999999"
> and the counter to be "99999". That would give
> "99,999,999,999,999" conbinations. We know that by
> manually setting the seed/counter, one can replicate
> a certain "sequence" of dice rolls. Let's say that
> an average game lasts 60-70 or even 100 rolls. You
> certainly would want to avoid repeating the same
> sequences and giving the player the impression of
> "deja-vue" too often. But what's the purpose of
> such large numbers? Are we supposed to somehow
> believe that JF can generate 999,999,999,999
> unique sequences of 100 dice rolls...? (which would
> require either a hell-of-an algorithm or crunching
> power of a "Cray" supercomputer).
No, it wouldn't. Generating an n-bit pseudo-random number from n bits of
state requires O(n) time using an ordinary linear congruential random
number generator (other methods might take longer, but a good linear
congruential generator is adequate for backgammon purposes anyway).
999,999,999,999 really isn't all that big -- it's under 40 bits. Taking
a quick look at the documentation for drand48() on my system, I see that
it can generate a 48-bit number from 48 bits of state with one multiplication
and one addition. This would take a microsecond at the outside. It hardly
needs a supercomputer! And drand48() isn't even a particularly good
generator. I see that under the random() function on my system (this is a
quote from the man page):
With a full 256 bytes of state information, the period of the random-number
generator is greater than 2^69, which should be sufficient for most pur-
poses.
2^69 32 bit integers contain 2^74 bits, and it require about 5.17 bits
to generate one dice roll (one pair of dice). So, it would be possible
to use this function to generate 3,653,000,000,000,000,000,000 rolls before
repeating the sequence!
> If there is no
> intention to eliminate repeating sequences, what's
> the big numbers good for? (other than perhaps an
> attempt to "make-believe"?).
To 32-bit arithmetic, the numbers Jellyfish uses really aren't very big at
all. ("Why did you use those numbers?" "Because they were there...")
> Personally, I think
> the number of sequences are finite, perhaps quite
> small and circular, with thousands of seed/counter
> combinations mapping to the same entry points.
I don't know how you can possibly believe this. Why are people so
suspicious of backgammon random number generators? Dice rolls for a
casual game of backgammon against a computer don't even _need_
particularly good random numbers; they only need to be sufficiently
random that the human can't predict what will be rolled next. There
are far, far more demanding applications for good random numbers in
the world (like cryptography). Given that there are programmers
capable of writing random number generators with _thousands_ of bits
of state for cryptographic purposes, why is it so hard to believe that
it's possible to write a generator with a few _dozen_ bits of state
that would be perfectly adequate for a backgammon game?
Patti Beadles
Dean Kezan <ke...@elender.hu> wrote:
>But than I remember one article where two champion's ( excellent player's)
>played with JF. It was bolded that they rolled manually dice's.
>What this proof ?
>That players was superstition on computer dice's or it is nicer to rolled
>manually.
I have no insight into why the gentlemen in question chose to play
with manual dice rather than computer-rolled. But I can make some
guesses. If I was playing for $200/point, I would certainly choose
to go with a technology that has been proven over the years, rather
than opening the contest up to any question of fair play.
>As I remember result was: one player won against JF and second loosed.
>This is also a proof that JF don't cheat on strategy or playing, which I
>never doubt.
>Am I right until now ?
Yep. Jellyfish broke exactly even on this contest. Nack Ballard won
a big pile of points, while Mike Senkiewicz (whose name I'm probably
misspelling) lost the same number. Given that Mike and Nack are
more-or-less evenly matched players, I consider this pretty conclusive
proof that Jellyfish plays somewhere in the same ballpark as them.
>But can someone prove even Mr. Author, that dice generator don't have
>anomalies in calculation ?
>Anomalies can be ( mathematically can be proofed) in poor dice formula so
>for an example: small numbers of yours can be big numbers of opposite ,or
>if you have big numbers it is also possible that you will get big numbers
>,or if you ... until the formula is "balanced" ( depend on type of
>formula).
To the best of my knowledge, no.
However, even if there are anomalies in the dice generator, it's
almost 100% certain that they're equally balanced, and don't provide
an edge to either player.
This is true of the backgammon servers too. If the servers have
accidental flaws in the dice generator, the flaws will be relatively
small, and they will occur uniformly, not favoring one player over
another.
So, if the question is, "Are the dice perfect?" I would have to say
that I don't know. "Are they fair?" The answer is a resounding yes.
-Patti
--
Patti Beadles |
pat...@netcom.com/pat...@gammon.com |
http://www.gammon.com/ | Try to relax
or just yell, "Hey, Patti!" | and enjoy the crisis
Claes Thornberg
>
> James Eibisch wrote:
>
> > Murat Kalinyaprak wrote:
>
> >>Just in case it was misunderstood, I wasn't proposing
> >>that JF gets itself out of "tough positions", etc. by
> >>rolling what it needs. JF's seed/counter scheme takes
> >>care of any such claims already. What I was proposing
> >>was that, it "leads" the game (and also the opponent)
> >>to that position.
>
> > <----------------- perfectly blocked ---------------->
>
> > Murat does it again! It must be deliberate, surely?
>
> In your newsreader program, click on preferences, set
> seed=5555/counter=222 and start typing. What you type
> will automatically self-justify... :)
>
> But seriously, yes it's deliberate. It's kind of like
> playing cross-word puzzles, word-games, etc. It slows
> down my typing a bit but also gives me another chance
> and more time to re-think what I'm saying...
>
> MK
>
>
Murat Kalinyaprak says that it's deliberate that his
posts are prefectly blocked, but I believe he/she is
a fraud, or a cheater. To my knowledge, no human can
possibly, on such a regular basis, write text, which
isn't in need of pad characters to become both right
and left justified. It is my belief that Kalinyaprak
is in league with some dark forces, or at least, has
foreknowledge of what words he will write. If that's
not cheating, I don't know what is. But to get on to
what really bothers me. Murat maintains that it also
gives him a second chance and more time to think and
re-think what he is saying. If so, how come that not
a single one of his posts, and I have read them with
great interest, contain any proof that Jellyfish can
predict the dice rolls and choose its plays based on
this knowledge? And before anyone starts this thread
again, yes I know it is held impossible to show that
Jellyfish doesn't cheat, and that there's conclusive
proof that it doesn't. So I will let Mr. Kalinyaprak
stick to his misconceptions about Jellyfish cheating
and Jellyfish being just another computer program he
can beat. All I can say, judging from some positions
in his matches against Jellyfish, is that he doesn't
seem to be such a good player himself. No player can
call himself good or knowledgable when redoubling in
a match when leading 13-0 to 15. And then it doesn't
matter how much perfectly blocked nonsense they post
to this newsgroup. Their opinions are not, and never
will be, taken seriously by anyone, except maybe the
well-known Mr. Andrade. But fortunately, he has now
earned himself a permanent position in my kill file.
Well, if writing this post wasn't just a waste of my
valuable time, my name isn't Claes Thornberg. But as
anyone who bothers to read through the .signature at
the bottom of this post would conjure, I call myself
Claes Thornberg, or my .signature is a fake. I would
like to express my sincere thanks to Mr. Kalinyaprak
for his entertaining posts about Jellyfish cheating,
and also for incredible ability to keep his posts so
perfectly blocked. I cannot understand how you could
possibly write posts like that. Either you're a kind
of linguistic genius, or have to much time to waste.
Be that as it may, I strongly believe you have a lot
of time to waste, and suggest you use it to become a
better backgammon player. Then you would feel little
less inclined to write such nonsense as your earlier
posts about JF cheating. Well, it seems that writing
perfectly blocked text is so catching, I haven't got
any idea how to stop. Fortunately, I have to go home
now. So, bye for now, and I hope we'll see some more
enlightning posts about how Jellyfish cheats. If you
are depressed, in a bad mood, or just bored, there's
nothing to cheer you up like another of those posts.
Regards, and excuse my drivel Murat, Claes Thornberg
--
____________________________________________________
Claes Thornberg Internet: cla...@it.kth.se
Dept. of Teleinformatics URL: No personal web-page!
KTH/Electrum 204 Intl phone: +46 8 752 1377
164 40 Kista, Sweden
Murat Kalinyaprak
>In <yvkvhos...@cuchulain.it.kth.se> Claes Thornberg wrote:
>>Murat Kalinyaprak <mu...@cyberport.net> writes:
>fonts. In this group, people must be using fixed
>fonts to illutrate/see board positions, and many
>readers have noticed it quickly...
>>and left justified. It is my belief that Kalinyaprak
>>is in league with some dark forces, or at least, has
>>foreknowledge of what words he will write. If that's
>>not cheating, I don't know what is.
>Heh, heh... Of course, I know what words I will write
>and make minor adjustments. (Just like Jellyfish?:)
PS: Did you notice that I also adjusted my line-length
to match yours from this point on? :)
MK
Murat Kalinyaprak
>Murat Kalinyaprak <mu...@cyberport.net> writes:
>> In your newsreader program, click on preferences, set
>> seed=5555/counter=222 and start typing. What you type
>> will automatically self-justify... :)
>Murat Kalinyaprak says that it's deliberate that his
>posts are prefectly blocked, but I believe he/she is
Backgammon comments to follow further below. But
first, let me say a couple of things about this.
I'm glad I may have started an interesting trend
in this newsgroup. I occasionally did this trick
in other newsgroups, but they haven't noticed it
much, perhaps because they mostly use proportial
fonts. In this group, people must be using fixed
fonts to illutrate/see board positions, and many
readers have noticed it quickly...
>a fraud, or a cheater. To my knowledge, no human can
>possibly, on such a regular basis, write text, which
>isn't in need of pad characters to become both right
>and left justified. It is my belief that Kalinyaprak
>is in league with some dark forces, or at least, has
>foreknowledge of what words he will write. If that's
>not cheating, I don't know what is.
Heh, heh... Of course, I know what words I will write
and make minor adjustments. (Just like Jellyfish?:)
>But to get on to
>what really bothers me. Murat maintains that it also
>gives him a second chance and more time to think and
>re-think what he is saying. If so, how come that not
>a single one of his posts, and I have read them with
>great interest, contain any proof that Jellyfish can
>predict the dice rolls and choose its plays based on
>this knowledge?
I haven't necessarily dwelled on a certain way as how
Jellyfish cheats, but it's correct that so far I have
provided nothing more than some speculations, and not
yet any concrete proof. However, I will recap some of
my previous propositions later in this article, which
you may want to at least consider mildly.
Next to the issue of cheating, other stuff like those
extremely large seed/counter numbers/combinations can
be "proven" to be a gimmick. This in itself is no big
problem, but I don't like gimmicks and feel justified
to look for more fishy stuff where I see one...
>proof that it doesn't. So I will let Mr. Kalinyaprak
>stick to his misconceptions about Jellyfish cheating
>and Jellyfish being just another computer program he
>can beat.
Well, what am I supposed to make of the fact that (at
least so far) I could beat Jellyfish 60% of the time?
That I'm all of a sudden a "world-class player"? Give
me break. I have no such *misconceptions*...
>All I can say, judging from some positions
>in his matches against Jellyfish, is that he doesn't
>seem to be such a good player himself.
This is a very interesting argument indeed. The other
day, I received an e-mail that briefly stated:
"I haven't seen you play. But I know from the
"arguments you make, that you can't be a very
"good backgammon player either.
I make some moves against Jellyfish. They work for me
consistently enough to keep me ahead so far. And then
you guys come tell me that I'm not a good player? How
do you know that my moves are not deliberate? What if
I'm smart enough to "read/figure out" my opponent and
adjust my play accordingly (which could be especially
easy against a *static* opponent as a computer game)?
In my previous articles, I had given some guesses and
hints as to how this could done. I'm still pondering
and experimenting on it and I may be able to provide
better ones yet later on...
>No player can
>call himself good or knowledgable when redoubling in
>a match when leading 13-0 to 15.
You have a point but I have a few reasons for it that
you may want to consider. I keep track of my score as
points, as well as one-point games. If the goal is to
score points, why not double and win more when I know
I can? Also, as a human player, there is an emotional
side of me that gets pissed at JF's doubling when (at
least in my personal opinion) I feel it shouldn't but
should gracefully drop (let alone it resigning, which
I'm not sure if it knows how to). Finally, I also get
tired of having to finish each game to the last move,
while having to stare at the "hour-glass" in between,
and I try to cut it short by forcing JF to drop (with
no such luck, of course:(. Still, I would never argue
that I know how to use the cube correctly. Neither do
I have the slightest idea on how to calculate equity,
etc. Only within the past few weeks that I started to
try making sense and making use out of the pip count.
A lot of the "lingo" used in this newsgroup eludes me
also. But none of these are essential to playing good
backgammon and even counting each game one point, I'm
still quite ahead of JF.
>I would
>like to express my sincere thanks to Mr. Kalinyaprak
>for his entertaining posts about Jellyfish cheating,
My pleasure. If you liked them, I'll be happy to post
more:).
>and also for incredible ability to keep his posts so
>perfectly blocked. I cannot understand how you could
>possibly write posts like that.
Could it be because of that very same reason? Nothing
concrete may come out of my suspicions/speculations.
In the meantime, I truely admire some peoples "faith"
in Jellyfish.
>Either you're a kind
>of linguistic genius, or have to much time to waste.
>Be that as it may, I strongly believe you have a lot
>of time to waste, and suggest you use it to become a
>better backgammon player.
Could it be that the two coincide? Think about it. It
takes a few minor tricky adjustments to justify text,
and although it may compromize on expressing yourself
in the best way (i.e. making the best bacgammon move)
it does accomplish the result you want. What if I can
recognize a pattern in JF's play, predict (accurately
enough) what it's getting ready for and preempt it by
making minor adjustments in my moves...?
>Then you would feel little
>less inclined to write such nonsense as your earlier
>posts about JF cheating.
I am just honestly relating what I believe to observe
and experience with JF. For example, I had previously
advanced my impression that I could predict JF's next
roll. I made a dry-run playing 10 games and trying to
predict some rolls. What I achieved was 7/10, *7/16*,
13/15, 11/14, 5/9, 7/8, 2/2, 5/10, 8/13 and 7/7. This
adds up to about 70% correct guesses (on a 1/2 odds).
I fell below 50% only in one game (as indicated), but
with "real" odds I would still score above 50%, since
one correct guess against 1/36 odds would allow me to
be wrong for 35 times before I fall below 50%. How do
you respond to things like this? Clearvoyance? Just a
temporary coincidence? I may not mind playing another
50 games and report the results, if you could tell me
ahead of time what would be your reaction/response if
the above ratio holds true even after several hundred
more guesses?
>Well, it seems that writing
>perfectly blocked text is so catching, I haven't got
>any idea how to stop. Fortunately, I have to go home
>now. So, bye for now, and I hope we'll see some more
>enlightning posts about how Jellyfish cheats.
Should the only acceptable proof be "technical"? What
about observations, statistics, game results, etc. as
"practical" proof...? I also feel that a "reasonable"
explanation which could come close to a technical one
may not be too far away. What if...?
MK
my_...@hotmail.com
mu...@cyberport.net (Murat Kalinyaprak) wrote:
>
> >No player can
> >call himself good or knowledgable when redoubling in
> >a match when leading 13-0 to 15.
>
> You have a point but I have a few reasons for it that
> you may want to consider. I keep track of my score as
> points, as well as one-point games. If the goal is to
> score points, why not double and win more when I know
> I can?
Hi Murat,
I think you're confusing your experiments a little.
When you play a 15 point match, the goal is to get to
15 points, and there can be no extra triumph in scoring
more. (i.e. by offering a double in positions Jellyfish
will never reject, such as when you are ahead 13-0 and
you redouble).
So, when you play 15 point matches, the only thing you
should be recording is how many 15 point matches you
have won.
If you want to keep statistics on points won, you
should choose "single game" (not match) and make sure
you disable the Jacoby rule, so that the cube can
be used normally.
If you want to record how well you do in terms of
games won (not affected by the cube), choose 1 point
matches.
> Finally, I also get
> tired of having to finish each game to the last move,
You can hit control-A to complete the game very quickly
at any time. (or hit the "stickman" button)
> A lot of the "lingo" used in this newsgroup eludes me
> also. But none of these are essential to playing good
> backgammon and even counting each game one point, I'm
> still quite ahead of JF.
>
I agree that you are ahead of JF, based on what you're
reporting, but I disagree that you can demonstrate that by
counting each game as one point. If that is the truest
indicator of skill, then you should be playing 1-point
matches. You will notice right away that Jellyfish
plays a different style when the goal is to win a
single point.
> In the meantime, I truely admire some peoples "faith"
> in Jellyfish.
People's "faith" as you put it ... is not exactly blind
faith. We've seen Jellyfish record an extremely high rating
on FIBS, and it played more than 6000 games in doing so. And
it certainly didn't roll it's own dice there.
Also, the real value of the product as a *tutor* comes from
it's ability to quantify your errors (in terms of lost equity)
and do *rollouts* of any position. The free download version of
the product doesn't give you those options.
No one says Jellyfish is perfect. In fact, some of the more
interesting discussions on this newsgroup come from people
questioning JF's decisions. So on that note, I propose that
if you want to convince people that Jellyfish's game is
flawed, please post checker plays or cube decisions it has
made which you feel are incorrect.
> I am just honestly relating what I believe to observe
> and experience with JF. For example, I had previously
> advanced my impression that I could predict JF's next
> roll. I made a dry-run playing 10 games and trying to
> predict some rolls. What I achieved was 7/10, *7/16*,
> 13/15, 11/14, 5/9, 7/8, 2/2, 5/10, 8/13 and 7/7. This
> adds up to about 70% correct guesses (on a 1/2 odds).
Can you explain why you feel your odds are 1/2 ?
Or perhaps clarify what exactly you're doing in this
experiment.
Murat Kalinyaprak
> MK> like Dejanews. I don't recall being accused of trolling
> MK> too often, if at all, and I probably have a pretty good
> MK> credibility in newsgroups where I had been posting on a
> MK> regular basis.
>Well it wasn't an accusation...I like trolls. :)
>......<----------------------------------------------------->
>But again, what's the secret?
Just use alternate words, like may/might, think/believe,
etc. and other tricks like an extra comma where it
wouldn't hurt. There is more than one way to express
the same idea, constructing slightly different sentences.
Once you get used to it, it's no big deal. Especially for
people who may be used to writing poems and rhyming with
synonyms, etc. Give it a try...
MK
Murat Kalinyaprak
>Your original comment, that Jellyfish tends to roll
>exactly what it needs, was disproven in spectacular
>fashion based on these two games.
I had read old articles about the seed/counter stuff
before my original comments. So I doubt that I said
it was rolling what it needs, but I did and still
think that it leads to such situations. I will use
an example game to illustrate this in a different
article. But for now, if you mean that JF's loosing
is a proof for its not cheating, I would disagree.
I can't demonstrate for sure that it cheats, but if
it does it can still be beaten despite it.
MK
Murat Kalinyaprak
>In <6oqv34$4to$1...@news.chatlink.com> Murat Kalinyaprak wrote:
>>Take me as an example of the ones who have been
>>"complaining here". Have you ever seen me play?
>>Do you know how good/bad I may be at positioning
>>my pieces? The answer is "no". Yet, you present
>>such an argument as a "fact"...?
>There's a story about a grand master chess player who is on
>a train and comes across a man with a chess board in front of
>him. He sits down and asks the man if he'd like to play a
>game.
>
>The man agrees, and asks what stakes the master would like to
>play for.
>
>"Any amount you choose", replies the master.
>
>"WHAT?!? Do you have any idea who I am?"
>
>"No, none at all. And that's why I'll let you choose the stakes."
Good story. You and folks like the guys who offered to
back JF at $10/point (or whatever) against me could/should
learn something from it. You guys know JF (idendify with
it) and just based on that knowledge, you pass baseless
generic judgements on other people whom you don't know
anything about. You are the eager ones to talk about
betting, stakes, etc. But I like your story anyway and
won't dwell on this any further...
MK
Claes Thornberg
>
[STORY DELETED]
>
> Good story. You and folks like the guys who offered to
> back JF at $10/point (or whatever) against me could/should
> learn something from it. You guys know JF (idendify with
> it) and just based on that knowledge, you pass baseless
> generic judgements on other people whom you don't know
> anything about. You are the eager ones to talk about
> betting, stakes, etc. But I like your story anyway and
> won't dwell on this any further...
>
> MK
Well, I guess that the people offering to back JF at $10/point would
be willing to back it for any amount you choose. There is, as far as
I know, at least one who is would be willing to do that. In previous
posts, he offered to pay for flight and hotel, only the person would
play at least a certain amount of games, the number of which I don't
remember. So, to me it's more the other way around. It's you who are
claiming that Jellyfish cheats and you can beat it with manual dice.
They don't know you but they know Jellyfish, and therefore they have
generously offered you to play for any amount you want. Moreover, as
for not knowing anything about you, you have supplied proofs to this
newsgroup about your knowledge of backgammon. I haven't analyzed the
matches yet, but judging from some positions from these and comments
that were made, I'd say that they are not likely to lose money. Take
this comment as you want, it is not my wish to ridicule your ability
to play backgammon. No matter who you were, they wouldn't be risking
losing much, in my opinion, but I won't dwell on this.
Regards,
Claes
--
______________________________________________________________________
Claes Thornberg Internet: cla...@it.kth.se
Dept. of Teleinformatics URL: NO WAY!
KTH/Electrum 204 Voice: +46 8 752 1377
164 40 Kista Fax: +46 8 751 1793
Sweden
Murat Kalinyaprak
>The algorithm is initialized with some number, called a seed.
>After the first random number is produced, that number is
>then used as part of the equation to get the next number.
>And so on, ad inifinitum. Eventually the algorithm will
>wrap around on itself, but for a 32-bit random number that's
>not going to happen very quickly.
I think you and a few others are missing the point on
this issue. The initial random number is irrelevant.
You can arrive at a random number by taking the number
of stars in the universe, subtract your social security
number, divide by your grandmothers birthday, etc. or
using much smaller values, depending on the scope. The
question here is whether JF's numbers are exaggerated
for their actual scope.
The fact that you can set the seed=1/counter=0 a
hundred times over and each time roll the exact
sequence of 99999 dice rolls means that there is
nothing random about it. If I'm given the impression
that I can set the seed=2/counter=0 and roll another
99999 sequences of dice rolls, I would expect that
the first set of 99999 rolls will not be exactly
the same as the second set of 99999. Now, that's
pretty easy for 2 sets. Even if 99998 rolls are
the same and only 99999th roll is different, that
would make each a unique set. But what about all
999,999,999 sets being all unique?
Furthermore, an average game seems to last about
maybe as little as 40 rolls. Let's say 100 rolls.
Now, if I choose seed/counter values 765/24, 59/8888,
27895245/12, 54545/2323, 769999200/1 and play 5 games,
you wouldn't want me to end up rolling the same exact
set of 100 rolls despite having chosen 5 different
seed/counter values each time, would you...? Ok, this
would require you to make 999,999,999,999 sets of 100
(99) sets of dice rolls unique and may be too much to
ask for (beside being mathematically impossible?). But
can you even make sure that the above mentioned
999,999,999 sets of 99999 dice rolls are all unique?
If you could post here an algorithm that could achive
such a feat with even 1000 sets of 100 sequences (and
not display the hourglass for several hours on an
average PC), I would indeed bow in front of you in
deep reverence...
Note that I'm only dwelling on the idea of algorithm
beacause that's how you claimed that those dice rolls
are generated and also because obviously they couldn't
be being fetched from fixed data stored in a file (since
that would have reqired all of us to have hard drives
of 99,999,999,999,999+ bytes capacity, or maybe just
half of that using another *algorithm* to compress it
also:)...
So, if the intention or the claim is not to produce
999,999,999 sets of unique sequences of 99999 rolls,
why bother with such astronomical numbers, perhaps
other than for giving players a phoney impression...?
And why break it in two, under two names (i.e. seed and
counter)? Just let me type a 14 digit number and you
figure out at what position in your great circular
set of rolls you want to start rolling from. From that
point on, just roll me dice and when you reach
100,000,000,000,000 wrap around and keep rolling...
Do you see why I'm dwelling on the uniqueness implied
by the "chunks" (i.e. "counters") of sequences...?
The same purpose can be accomplished with much fewer
sets and smaller numbers. Let's say your entire set
consists of 500 rolls. After 10 games lasting 55, 42,
38, 57, 89, 22, etc. rolls, the 11th game may start
on the 486th roll and wrap around, after that, the
next 10 games may start on the 44th roll and last 60,
39, 75, 39, 18, etc. rolls. Even if one manually sets
the seed (i.e. from 1-10) and counter (i.e. from 1-50)
for each game, the chances of getting exactly the same
X-roll sequences often enough to be noticed would be
already very small with a well "refined" set of 500.
As far as randomness and fairness is concerned, this
could do the job just fine as the repeating rolls
will have different usability in different games or
even at different stages of the same game. As far as
avoiding players to get "deja-vue" too often, it
should be good enough also, since no average human
can possibly memorize several sequences and dozens
of dice rolls. I would be impressed if somebody
could notice that a certain sequence of 5 rolls had
happened in another game, even if they happened only
a few minutes apart.
If an algorithm generating *non-random* sequences, has
a repeat-rate of X, that's not going to change or get
any better by using smaller or larger numbers. What may
be JF's repeat rate? And if that repeat rate is not going
to get any more visible by reducing the counter/seed
sizes to X digits, why not use smaller digits instead...?
I would guess that a set of 10,000 rolls (i.e. 100x100)
or 100,000 rolls (i.e. 100x1000) should certainly be
more than enough. So, to anybody who understands what
I'm talking about and would care to offer comments,
the question remains: "What is the purpose of JF's
allowing such large numbers for seed/counter values?
And if you think it's a "make-believe", what do you
think the real scope of JF's dice roll sequences?".
MK
PS: If you set counter to 99990 and start rolling,
after 9 rolls it goes on to display 100000, 100001...
Claes Thornberg
>
> You have a point but I have a few reasons for it that
> you may want to consider. I keep track of my score as
> points, as well as one-point games. If the goal is to
> score points, why not double and win more when I know
> I can?
If you like to keep score, why not just play money-game
which I believe would be more sensible. And some of the
drops you'd expected maybe would have occured. I've yet
to see anyone who is used to playing tournament matches
minding losing 32-0 in 17 point match. However, there's
one reason for not getting any point at all in a match.
Sometimes I play 5 point matches with the addition that
if you "blitz" your opponent, i.e. you win 5-0, you get
paid double. But this is a bit off topic.
> Also, as a human player, there is an emotional
> side of me that gets pissed at JF's doubling when (at
> least in my personal opinion) I feel it shouldn't but
> should gracefully drop (let alone it resigning, which
> I'm not sure if it knows how to).
If you think Jellyfish should drop, and even gracefully
so, you should rejoice when it takes your double. I can
add that if an opponent of mine takes when I think it's
a drop, I get very happy, since I will win even more by
such a mistake.
> Finally, I also get
> tired of having to finish each game to the last move,
> while having to stare at the "hour-glass" in between,
> and I try to cut it short by forcing JF to drop (with
> no such luck, of course:
In a game where you are 100% sure of winning, you could
let Jellyfish play the game to conclusion on its own. I
usually do this myself, and recommend you do it to. You
can do this by pressing the button in the control panel
which looks like a running, or a walking man. Jellyfish
will then play both sides, using level 5.
Murat Kalinyaprak
>mu...@cyberport.net Murat Kalinyaprak writes:
>> articles specificly about JF. In all those, only
>> the argument that JF plays a given board always
>> the same way deserves any merit. However, that
>> argument goes against JF, in that it effectively
>> defeats any claims about JF's using neural/fuzzy/etc.
>> logic.
>That's not true at all. It uses a neural net evaluation
>function but not fuzzy logic. Playing the same way for
>a given position is exactly what you would expect from a
>neural net (given that it has already been fully trained
>and so its weights never change)
It sounds like your are saying that neural evaluation
was used prior to/until things were cast in conrete...?
How can there be such a thing as "fully trained" when
luck is involved? "Fully trained" to do what? The best
move in each situation? And what's the criteria for the
best moves? That they produced the best results over a
100,000 games? But backgammon depends on dice and after
another 200,000 games, what would "neural evaluation"
would consider "best moves" may change...
My understanding of "neural" applied to computer programs
is that they are capable of self-adjusting/self-improving
in a dynamic way. I don't think subsequent static usage
of some neural evaluation results qualify JF as "neural".
In my opinion/understanding, there is more to neural than
just static logic built merely upon large statistics.
Anyway, it doesn't matter what it's called. JF's moves
are (or seem to be) cast in concrete...
>> Since what one observes doesn't quite jive
>> with what would be considered "random" by some
>> people, one keeps going back to questioning this
>> seed/counter scheme and whether there may be any
>> correlation between it and board positions, etc.
>How can there be? You can set the seed and counter manually,
>and you can change the board position too. They are utterly
>independent.
With one backgammon program I had a very strong
impression that the dice sequences were pre-recorded
real moves. Of course, in that program one had no
control of dice rolls, etc. and such a sceme would
be very easy to implement. In JF, skillfully arranged
and reciprocally difficult sequences (even if they are
short) may be dispersed among the dice roll sequences.
Whichever player can deal with them better (and this
may mean NOT making the best/expected move) may have
the necessary slight edge. One or two such positions
is all that's needed to effect the game's outcome.
I admit that my speculations sound too far-fetched
even to myself (mostly because implementing such
schemes would probably be a nearly impossible task).
But, what I think I observe keeps me pondering on it.
I'll post an actual game with JF rolled dice as an
example, in a different article. If you get to read
it, let me know what you think of it.
JF's making the same move in the same board positions
is not the issue for me. I have a feeling that the
dice sequences are leading to decisive positions, not
necessarily succeeding in doing it all the time (in
which cases JF would have no control over who wins).
But if a small number of such cases succeed, it may
be enough to give it an edge.
Let me also clarify that I never questioned whether
JF plays well. It does an impressive job even with
manual dice rolls (compared to many other similar
programs I;ve seen). The small intrigue (for the ones
who has nothing better to do) is whether it cheats
at all or not. In fact, if some of what I speculate
about could be implemented at all, I may admire it
as a "strategy" and stop using the word "cheating"
for it...
>> Personally, I think
>> the number of sequences are finite, perhaps quite
>> small and circular, with thousands of seed/counter
>> combinations mapping to the same entry points.
>I don't know how you can possibly believe this. Why are
>people so suspicious of backgammon random number generators?
>....
I think you missed the issue I was raising. Please
see to my response to another writer on this subject.
MK
Claes Thornberg
>
> In article <6p0gor$ug$2...@news.chatlink.com>,
> mu...@cyberport.net (Murat Kalinyaprak) wrote:
>
>
> > In the meantime, I truely admire some peoples "faith"
> > in Jellyfish.
>
> People's "faith" as you put it ... is not exactly blind
> faith. We've seen Jellyfish record an extremely high rating
> on FIBS, and it played more than 6000 games in doing so. And
> it certainly didn't roll it's own dice there.
>
A good point. Many people have faith in things without
seeing any proofs that they are true. And now, when we
have suggestive proofs that Jellyfish plays backgammon
quite well, not to say excellent, then of course, will
people believe anything but that. All I can say is, it
must be very frustrating having written one fine piece
of software and then have people accusing you of being
a cheater. Or it is extremely funny, depending on your
sense of humor.
Murat Kalinyaprak
>In <6p0gor$ug$2...@news.chatlink.com> Murat Kalinyaprak wrote:
>> You have a point but I have a few reasons for it that
>> you may want to consider. I keep track of my score as
>> points, as well as one-point games. If the goal is to
>> score points, why not double and win more when I know
>> I can?
>Hi Murat,
>I think you're confusing your experiments a little.
I think you are right and thanks for explaining the
differences between matches and single games, etc.
Initially I had posted the results of 3 matches as:
ME JF
-- --
25 6 (from 13-6)
40 6 (from 8-6)
20 0 (from 4-0)
I didn't think that anything other than winning the
match mattered. But I felt a need to at least note
the score just before the last game in each match,
mostly because each last game was a high-points
game and I had also wanted to make some comments
related to it.
In response to it, somebody indicated that a few
matches lasting only a few games wouldn't mean
much more than perhaps my being lucky and running
up the dice. After that, I posted 10 more match
result and itemized the games and evaluated the
results in terms of games because of this. I'm
not disputing what you said; I'm just explaining
how it got to my getting a little confused about
what mattered.
>> Finally, I also get
>> tired of having to finish each game to the last move,
>You can hit control-A to complete the game very quickly
>at any time. (or hit the "stickman" button)
Thanks for the tip. I may use it in situations
like during unthreatened bearoffs, etc.
>> A lot of the "lingo" used in this newsgroup eludes me
>> also. But none of these are essential to playing good
>> backgammon and even counting each game one point, I'm
>> still quite ahead of JF.
>I agree that you are ahead of JF, based on what you're
>reporting, but I disagree that you can demonstrate that by
>counting each game as one point. If that is the truest
>indicator of skill, then you should be playing 1-point
>matches. You will notice right away that Jellyfish
>plays a different style when the goal is to win a
>single point.
I sure noticed the difference in style. The games
seem to be shorter and JF seems to think not much
longer even though I upped time factor from 100
to 1000 (except very unnecessarily in some weird
situations, like you have 4 pieces left to pick, it
has 7 pieces all over the board with none being a
threat to you, and it just thinks and thinks and
thinks, perhaps 15-20 times longer than it would in
a much more complicated position). BTW: I set the
time factor to 1000 because it seems to be the
highest value JF allows for it. Is this true and
how critical is it to the difficulty level?
Currently, I'm in the process of playing 100 games
of 1 point each, with automatic dice roll. I'll
post the results when I get done playing them.
In the meantime, would anybody care to make predictions
on the (expected) outcome...?
>> In the meantime, I truely admire some peoples "faith"
>> in Jellyfish.
>People's "faith" as you put it ... is not exactly blind
>faith. We've seen Jellyfish record an extremely high rating
>on FIBS, and it played more than 6000 games in doing so. And
>it certainly didn't roll it's own dice there.
I don't play on fibs, nor am I rated by other means.
If I were to tie with JF, what would be my "rating"?
>No one says Jellyfish is perfect. In fact, some of the more
>interesting discussions on this newsgroup come from people
>questioning JF's decisions.
Since I have done a lot of that and almost nothing but
that, here I'll offer a few constructive suggestions
that would make JF more valuable in my opinion:
1- Record the seed/counter in the mat/gam/pos files
and reset them to those values when such files
are reloaded. That way, one can re-play the same
game/match without having to jot down and later
manually reset seed/counter values.
2- Allow a selection to roll dice automaticly from
a chosen gam file. This way, if a game was played
with manual dice rolling, it can be re-played as
explained above (the difference being that the dice
rolls would be predetermined by a previous game
instead of a seed/counter pair).
3- During automatic dice rolls, allow skipping a roll
for either player, consistently and in a predetermined
interval. This way one can play 2 out of every 3 rolls,
etc. and may feel better about having broken any
pre-determined patterns in the dice sequences... :)
>> I am just honestly relating what I believe to observe
>> and experience with JF. For example, I had previously
>> advanced my impression that I could predict JF's next
>> roll. I made a dry-run playing 10 games and trying to
>> predict some rolls. What I achieved was 7/10, *7/16*,
>> 13/15, 11/14, 5/9, 7/8, 2/2, 5/10, 8/13 and 7/7. This
>> adds up to about 70% correct guesses (on a 1/2 odds).
>Can you explain why you feel your odds are 1/2 ?
>Or perhaps clarify what exactly you're doing in this
>experiment.
I meant to say "*even* at 1/2 odds", as it can't get
lower than that for any type of guessing. Why do such
experiments? Because I think I observe something and
I hear other people express similar impressions of
theirs. Then I see some people arguing back that it
just seems that way to some people (even referring to
them as "lesser players", etc.). And I think this is
one way to substanciate one's impressions/observations.
100 guesses in 10 games may not be quite enough, but
what if one can guess the next roll while playing
against JF, correctly 70% of the time after 2,000
guesses in 200 games...? What would you say then?
Would you tend to believe that the person must be a
psychic or would you start suspecting JF's dice rolls...?
MK
Murat Kalinyaprak
>Also, if it is your belief that jellyfish cheats, you must
>also think that it is making incorrect moves. Can you post
>examples of what you consider a Jellyfish error? (and include
>the match length and score at the time)
I don't want to get into arguing on dozens of moves
as to whether they were the correct/incorrect ones.
But here is a game that may illustrate how one can
beat JF by making the "incorrect" (in my opinion)
moves.
I'll re-play the game as I write, and make comments
at certain positions. I'll also attach the gam file
intact at the end, so that you can cut and paste it
into a gam file and play it.
10th game of a match - JF=13 MK=10 - Seed=3 Counter=818
Nothing much unusual at the beginning. MK doubles at
6th move and JF takes. After that, you may start
seeing MK make some "incorrect"? moves, like the 63
at 9th move. Now watch closely after JF's 11th move,
where it spreads further to hit me.
11) MK-43: Of course, I hop over without hitting
12) MK-32: I hit and cover, JF get's back in
13) MK-63: Look how I play. Is this how you would
play this one? Isn't this the "incorrect"
move? But it works! JF doesn't hit neither
of my pieces, while it would have hit one
if I had played "correctly".
14) MK-43: Two more of my pieces are now safely away.
15) MK-61: I have no choice but leave two more pieces
open. Playing correctly, you should separate
them by more than 6 positions (if you can)
so that they can't be both hit by one single
piece of your opponent in a situation like
this one, right? But I play the other way,
and again it works! JF hits neither of my
pieces, while it would have hit one if I had
played "correctly".
16) MK-43: Move at least one piece to safety and leave
only one open, right? Nope. Do the opposite.
And by golly, it works once more! Neither of
my pieces get hit, while one would have been
hit if I had played "correctly". Look at JF's
board during these last moves, and you can
see how critical it was for me not to get hit.
One hit would have been most likely the end
of the story for me.
17) MK-54: My last two pieces are safely home. It's all
over for JF...
It may be plain imagination/superstition or whatever
else you may want to call it, but it worked for me in
this situation, as similar moves worked in other
situations. In this case it all starts with how I play
63 in the 9th move as I mentioned above. Had I played
that and the 53 in the 10th move "correctly", I wouldn't
be able to play the 11th move and the ones followed it
the way I did. I don't know what most people would think
of such a sequence, but this many "incorrect" moves in
a row saving me again and again, while just one "correct"
move instead of any one of the "incorrect" ones would
have ensured my getting hit, is a little past simple
coincidence. Especially if similar sequences seem to
happen again and again...
Even if baseless, it's at least fun trying to beat JF
at "its own game" by such tactiques...
Below is the complete gam file. I hope you have fun with
it too.
MK
--------------------------------------------------------
JellyFish Player2
1) 43: 13/9 24/21 62: 24/16*
2) 52: 25/23 13/8 63: 16/10 13/10
3) 31: 8/5 6/5 33: 10/7 10/7 6/3 6/3
4) 42: 8/4 6/4 11: 8/7 6/4* 4/3
5) 32: 25/23 24/21 65: 24/13
6) 22: 23/21 23/21 13/9 Doubles => 2
7) Takes 21: 3/1 7/6
8) 65: 21/10 55: 13/3 13/8 6/1
9) 21: 9/7 8/7 63: 13/7 6/3
10) 55: 10/5 8/3 7/2 7/2 53: 7/2 6/3
11) 11: 21/19 5/3 43: 8/1
12) 31: 19/16 6/5 32: 7/4* 4/2
13) 42: 25/21 16/14 63: 8/2 8/5
14) 64: 14/4 43: 7/3 5/2
15) 62: 13/7 13/11 61: 13/7 13/12
16) 42: 11/7 5/3 43: 12/5
17) 52: 7/2 3/1 54: 7/3 5/0
18) 31: 21/18 7/6 21: 2/0 1/0
19) 41: 18/13 53: 3/0 3/0
20) 22: 13/5 51: 3/0 1/0
21) 62: 6/0 2/0 52: 3/0 2/0
22) 42: 4/0 2/0 52: 3/0 2/0
23) 61: 6/0 1/0 42: 3/0 2/0
24) 43: 4/0 3/0 61: 3/0 1/0
Wins 2 points
--------------------------------------------------------
Patti Beadles
Murat Kalinyaprak <mu...@cyberport.net> wrote:
>It sounds like your are saying that neural evaluation
>was used prior to/until things were cast in conrete...?
>How can there be such a thing as "fully trained" when
>luck is involved? "Fully trained" to do what? The best
>move in each situation? And what's the criteria for the
>best moves? That they produced the best results over a
>100,000 games? But backgammon depends on dice and after
>another 200,000 games, what would "neural evaluation"
>would consider "best moves" may change...
>My understanding of "neural" applied to computer programs
>is that they are capable of self-adjusting/self-improving
>in a dynamic way. I don't think subsequent static usage
>of some neural evaluation results qualify JF as "neural".
>In my opinion/understanding, there is more to neural than
>just static logic built merely upon large statistics.
You're both right and wrong.
Neural networks have two modes. In one mode they're learning
and adjusting the weights as they go. In the other mode, they
always produce the exact same output given the same input (in
the case of backgammon, a position.)
Neural networks are trained by having them play a huge number
of games (millions, I believe) against usually a fixed opponent...
one that odes the same thing in every situation. And during
this time, what the net considers the best move will indeed
change as it learns.
In the case of Jellyfish (and almost certainly Snowie), the net
has already been trained, and is now used only to evaluate
positions. It's no longer learning after it's been shipped to
you. And yes, that does still qualify it to be a neural net.
-Patti
--
Patti Beadles | Not just your average purple-haired
pat...@netcom.com/pat...@gammon.com | degenerate gambling adrenaline
http://www.gammon.com/ | junkie software geek leatherbyke
or just yell, "Hey, Patti!" | nethead biker.
Patti Beadles
Murat Kalinyaprak <mu...@cyberport.net> wrote:
>Good story. You and folks like the guys who offered to
>back JF at $10/point (or whatever) against me could/should
>learn something from it. You guys know JF (idendify with
>it) and just based on that knowledge, you pass baseless
>generic judgements on other people whom you don't know
>anything about.
I think you're missing my point.
If you were one of the (very few) people on this planet who
I think would be a favorite against Jellyfish, I'd probably
know who you are.
Gary Wong
> The fact that you can set the seed=1/counter=0 a
> hundred times over and each time roll the exact
> sequence of 99999 dice rolls means that there is
> nothing random about it. If I'm given the impression
> that I can set the seed=2/counter=0 and roll another
> 99999 sequences of dice rolls, I would expect that
> the first set of 99999 rolls will not be exactly
> the same as the second set of 99999. Now, that's
> pretty easy for 2 sets. Even if 99998 rolls are
> the same and only 99999th roll is different, that
> would make each a unique set. But what about all
> 999,999,999 sets being all unique?
They almost certainly are unique, but there's nothing amazing about
that. A large number of unique subsequences is neither necessary nor
sufficient for a sequence to be `good' for backgammon purposes. See
an earlier post of mine (Message-ID: <ieyn2ex...@cs20.cs.auckland.ac.nz>)
for more on what backgammon requires from a pseudo-random number generator.
> Furthermore, an average game seems to last about
> maybe as little as 40 rolls. Let's say 100 rolls.
> Now, if I choose seed/counter values 765/24, 59/8888,
> 27895245/12, 54545/2323, 769999200/1 and play 5 games,
> you wouldn't want me to end up rolling the same exact
> set of 100 rolls despite having chosen 5 different
> seed/counter values each time, would you...? Ok, this
> would require you to make 999,999,999,999 sets of 100
> (99) sets of dice rolls unique and may be too much to
> ask for (beside being mathematically impossible?). But
> can you even make sure that the above mentioned
> 999,999,999 sets of 99999 dice rolls are all unique?
>
> If you could post here an algorithm that could achive
> such a feat with even 1000 sets of 100 sequences (and
> not display the hourglass for several hours on an
> average PC), I would indeed bow in front of you in
> deep reverence...
Certainly, I can do that in one line (from the article referred to above):
n' = ( n x 1,103,515,245 + 12,345 ) mod 2^32
ie. if you take any integer (let's start with 1), and continually
multiply it by 1,103,515,245 and add 12,345 in modulo 2^32 arithmetic,
then you won't get the same number again until you've done 2^32
multiplications. That's a sequence of 4 billion unique integers. (I
tried it just now to make sure -- it really did take 4 billion
iterations before `1' appeared again, and it took just under 9 minutes
on an Alpha Station 200).
No need to bow before me in deep reverence. I didn't invent it anyway :-)
> So, if the intention or the claim is not to produce
> 999,999,999 sets of unique sequences of 99999 rolls,
> why bother with such astronomical numbers, perhaps
> other than for giving players a phoney impression...?
Nobody (except you) said anything about 999,999,999 unique sequences.
The point is that the internal random number generator has a certain
amount of state -- let's assume it's 32 bits (the same as the example
above). There are then around 4 billion unique states the generator
could be in (ie. whatever the current n is in the same example). Why
would you want to impose any limit smaller than 4 billion on the number of
states the user can select from?
> And why break it in two, under two names (i.e. seed and
> counter)?
No need. Just for convenience.
> Just let me type a 14 digit number and you
> figure out at what position in your great circular
> set of rolls you want to start rolling from. From that
> point on, just roll me dice and when you reach
> 100,000,000,000,000 wrap around and keep rolling...
> Do you see why I'm dwelling on the uniqueness implied
> by the "chunks" (i.e. "counters") of sequences...?
No, I don't! Uniqueness of _subsequences_ isn't necessary or sufficient
for the sequence as a whole to be `random'. Besides, if you look at small
enough subsequences in a large enough sequence, then they won't be unique
anyway (since there are a finite number of possible subsequences of a
given length, and if you take a large enough sequence then eventually
you'll exhaust them all).
> If an algorithm generating *non-random* sequences, has
> a repeat-rate of X, that's not going to change or get
> any better by using smaller or larger numbers. What may
> be JF's repeat rate? And if that repeat rate is not going
> to get any more visible by reducing the counter/seed
> sizes to X digits, why not use smaller digits instead...?
By `repeat rate', I assume you mean period. The period of the example
generator I gave above is 2^32, and that's a pretty poor random number
generator (though it's adequate for backgammon). I have absolutely no
doubt that Jellyfish's generator is at least as good, and in all
probability its period is 2^32 or more. So, 9 decimal digits _are_
required to make the entire period `visible'.
> I would guess that a set of 10,000 rolls (i.e. 100x100)
> or 100,000 rolls (i.e. 100x1000) should certainly be
> more than enough. So, to anybody who understands what
> I'm talking about and would care to offer comments,
> the question remains: "What is the purpose of JF's
> allowing such large numbers for seed/counter values?
> And if you think it's a "make-believe", what do you
> think the real scope of JF's dice roll sequences?".
You already answered your own question: because if you have a large
`state space' (ie. period), then you might as well provide a
correspondingly large seed/counter space. There is no make-believe.
I have been trying to kill these threads for weeks now without any success :-)
I think we've digressed too far and lost the point of the original accusation.
The claim is that Jellyfish `cheats' with some (unspecified) collusion
between its dice generator and its choice of play (ie. evaluation function),
correct?
Well, I can prove that it _doesn't_ cheat with only two assumptions. If it
does cheat, then one of my assumptions must be wrong. Please give me a
counterexample demonstrating which one is unjustified. The assumtions are:
1) The dice rolls depend ONLY on the seed/counter and NOTHING else (not the
board position; not the match score; not the phase of the moon).
2) The move Jellyfish selects is strictly that which is evaluated with
the highest equity; the evaluation may depend on board position, cube
value and position, match score, but NOTHING else (not the current or
future dice, whether manual or computer generated; not what you ate
for breakfast this morning; nothing).
If these assumptions are true, then you can see that the dice and moves
are utterly independent. You can set the seed/counter to whatever you like
at the beginning of the game. Any board position is just as likely to occur
in that game as it is in another game with random seed/counter values.
Jellyfish will play the same move in each case. With some seed/counter
values, the subsequent rolls will be good for Jellyfish; with other values,
they won't. But the dice cannot possibly affect the moves (by assumption
2), and the moves cannot possibly affect the dice (by assumption 1); therefore
Jellyfish does not cheat.
Assumptions 1) and 2) are both falsifiable. I cannot prove them (that
would take an exhaustive search of the entire seed/counter/board
position space, which is far too large). But I hypothesise that they
are true, and so far there has been absolutely no evidence to refute
either of them. This has led me to believe that both assumptions are
correct. I will continue to believe in their truth (from which
Jellyfish's honesty directly follows) until evidence is provided to
the contrary.
Gary Wong
> In <wtaf66b...@brigantine.CS.Arizona.EDU> Gary Wong wrote:
> >mu...@cyberport.net Murat Kalinyaprak writes:
> >> the argument that JF plays a given board always
> >> the same way deserves any merit. However, that
> >> argument goes against JF, in that it effectively
> >> defeats any claims about JF's using neural/fuzzy/etc.
> >> logic.
>
> >That's not true at all. It uses a neural net evaluation
> >function but not fuzzy logic. Playing the same way for
> >a given position is exactly what you would expect from a
> >neural net (given that it has already been fully trained
> >and so its weights never change)
>
> is that they are capable of self-adjusting/self-improving
> in a dynamic way.
During training, that is true. But the evaluator in the Jellyfish you
play against has already been fully trained and no longer `learns'.
> I don't think subsequent static usage
> of some neural evaluation results qualify JF as "neural".
> In my opinion/understanding, there is more to neural than
> just static logic built merely upon large statistics.
You could argue that there is or there isn't, it's really only a matter
of taxonomy. For a complete description of the internals of neural net
evaluation functions, I recommend you read all the papers on the subject
at:
http://forum.swarthmore.edu/~jay/learn-game/systems/gammon.html
theodore hwa
:
: PS: If you set counter to 99990 and start rolling,
: after 9 rolls it goes on to display 100000, 100001...
This just shows that the algorithm is probably one that takes a pair (m,n)
of arbitrary positive integers as a seed and generates a random number
sequence from it. Of course there exist m and n such that the first 100 (or
whatever) rolls agree. 99999 is just a limit in the input fields, not for
the actual random number generator.
Since we're talking about a random number generator, we don't know
what values of m and n collisions will occur. In particular, there could
be two small pairs for which the first 100 rolls agree. You suggest that
the maximum size of the seed and counter fields be limited because of
collisions, but for all you know the algorithm could agree on the first
100 rolls for (5,3) and (6,1) [just to take an example]. Limiting the
values doesn't necessarily help. Your space of seeds has to be much larger
than your space of roll sequences (assuming , as you do, that you have
some practical limit on the length of a BG game in mind.) There is no
"uniqueness" condition on an RNG, such as the sequences of the "first" x
seeds must be different, or whatever. That would be difficult and
unnecessary to design for an RNG.
The size of the input fields of the "seed" and "counter"
boxes are just arbitrary choices. They are not a reflection of "limits" of
the algorithm of the RNG. The generator is probably not something specific to
base 10 so there is no reason to think that it should stop at some
number of the form 9999....9.
Of course, this is not to say that there is not some limit to the RNG;
obviously its state must eventually repeat. Even 'simple' RNGs can have
many states and large periods, but the sequences do not come in
groups based on the size of the seed.
John Goodwin
Kalinyaprak) wrote:
>In <pattibEw...@netcom.com> Patti Beadles wrote:
>
>>In <6oqv34$4to$1...@news.chatlink.com> Murat Kalinyaprak wrote:
>
>>>Take me as an example of the ones who have been
>>>"complaining here". Have you ever seen me play?
>>>Do you know how good/bad I may be at positioning
>>>my pieces? The answer is "no". Yet, you present
>>>such an argument as a "fact"...?
>
>>There's a story about a grand master chess player who is on
>>a train and comes across a man with a chess board in front of
>>him. He sits down and asks the man if he'd like to play a
>>game.
>>
>>The man agrees, and asks what stakes the master would like to
>>play for.
>>
>>"Any amount you choose", replies the master.
>>
>>"WHAT?!? Do you have any idea who I am?"
>>"No, none at all. And that's why I'll let you choose the stakes."
I was hoping someone else would ask, but as they haven't I'll have to
admit my stupidity and do so.
What does this tale mean?
I would normally expect it to be followed by the grand master getting
beaten, and a point being made about taking your superiority for
granted.
As it is, it doesn't make much sense. The grand master must be fairly
stupid if he makes wagers *because* he has less information than he
might have.
OK, I've got a damp cloth ready to wipe the egg from my face.
JG
Charles/Andy Sachrajda
: >>"Any amount you choose", replies the master.
: >>
: >>"WHAT?!? Do you have any idea who I am?"
: >>"No, none at all. And that's why I'll let you choose the stakes."
: I was hoping someone else would ask, but as they haven't I'll have to
: admit my stupidity and do so.
: What does this tale mean?
I think the point is that is the first gentelman where someone who
had any chance of beating our grandmaster friend, then he would be known to
the grandmaster. This may not be through now a days, when the grandmaster title
has been somewhat devalued, and the technic of the average master has
considerably improuved, but at the start of the century, it was proably true
enough, and even these days, for a Grandmaster to lose to a player he has
not recognised would be fairly rare.
Charles
: OK, I've got a damp cloth ready to wipe the egg from my face.
: JG
Donald Kahn
Goodwin) wrote:
>On Wed, 22 Jul 1998 05:51:55 GMT, mu...@cyberport.net (Murat
>Kalinyaprak) wrote:
>
>>In <pattibEw...@netcom.com> Patti Beadles wrote:
>>>There's a story about a grand master chess player who is on
>>>a train and comes across a man with a chess board in front of
>>>him. He sits down and asks the man if he'd like to play a
>>>game.
>>>
>>>The man agrees, and asks what stakes the master would like to
>>>play for.
>>>
>>>"Any amount you choose", replies the master.
>>>
>>>"WHAT?!? Do you have any idea who I am?"
>
>>>"No, none at all. And that's why I'll let you choose the stakes."
>
>
>I was hoping someone else would ask, but as they haven't I'll have to
>admit my stupidity and do so.
>
>What does this tale mean?
>
The apparent point is: if the stranger were a player the master needed
to worry about, he would know his face. That might have been true 100
years ago, but perhaps not a prudent assumption in 1998.
DK
Murat Kalinyaprak
>On Wed, 22 Jul 1998 09:47:38 GMT, mu...@cyberport.net (Murat
>Kalinyaprak) wrote:
>>Currently, I'm in the process of playing 100 games
>>of 1 point each, with automatic dice roll. I'll
>>post the results when I get done playing them.
>>In the meantime, would anybody care to make predictions
>>on the (expected) outcome...?
>What version of JF are you using, and what level is it
>playing at?
Player 3.01 - 32bit version, set at level 7 - time
factor 1000.
>If you win 60% of 1-pt matches (cubeless and gammonless) against
>JF 2 or 3 at level 5 or above I'd say that's pretty good. Real
>experts may expect to win more than that on level 5, I'm not
>sure - comments anyone? Jellyfish playing at level 5 was rated
>around 1900 on FIBS.
Thanks for the feedback. Feel free to revise your
predictions for level 7 if you wish.
>>I don't play on fibs, nor am I rated by other means.
>>If I were to tie with JF, what would be my "rating"?
>Let's set you up as a player with a rating of 1500 and
>experience of 400 (i.e. you've gone through the high ratings
>change period). Let's give Jellyfish a rating of 2000 - near
>enough - and experience of 400. You're both "mature" FIBS
>players with a 500-pt rating difference.
>If you play 100 1-pt matches against JF and win one, lose one,
>win one, lose one, etc., your rating will be 1549.
>..................
Thanks for taking time to explain all this too. It
has no practical use for me to know my "rating",
but since everybody talks about their ratings, I was
just curious to know how it's arrived at and what
would it have been for me.
>I look forward to seeing the result of your 100 1-pt matches
>against JF. Please remember that a 1-pt match is cubeless; there
>seems to be some confusion over that in your posts so far. Apart
>from my own records, I don't think I've seen the results of any
>reasonably long series of games here, so it'll be educational.
I'll post the results along with some statistics
and comments. I'll also make gam files available
if anyone cares to look at them, evaluate them,
etc. I'm getting more familiar with JF's features
as I go on, and one thing I'll do, for example, is
to have JF auto-play each game starting at the
same seed/counter values, to see how well it would
have done against itself if it was in my seat. I'll
post those two sets of results side by side.
Would anyone care to comment on what kinds of
statistics would be meaningful/beneficial...?
MK
Murat Kalinyaprak
So...?
>(I tried it just now to make sure -- it really did take 4 billion
>iterations before `1' appeared again, and it took just under 9
>minutes on an Alpha Station 200).
In this case, maybe I should up my estimates of how
long a PC would display the hour-glass *for the task
I was talking about*...
>No need to bow before me in deep reverence. I didn't invent it
>anyway :-)
Not quite yet. It looks like we are still talking
about two different things. See below...
>> So, if the intention or the claim is not to produce
>> 999,999,999 sets of unique sequences of 99999 rolls,
>> why bother with such astronomical numbers, perhaps
>> other than for giving players a phoney impression...?
>Nobody (except you) said anything about 999,999,999 unique
>sequences. The point is that the internal random number
>generator has a certain amount of state -- let's assume it's
>32 bits (the same as the example above). There are then
>around 4 billion unique states the generator could be in (ie.
>whatever the current n is in the same example). Why would you
>want to impose any limit smaller than 4 billion on the number
>of states the user can select from?
What your and other writers' insisting on this random
number generation stuff are missing is that we don't
play backgammon with integer values from 1 to N billion.
For example, from an arbitrary location in your set
of 4 billion unique imtegers, let's take a subset of
a few numbers:
.. 734, 156956, 2, 3456789, 888888, 1900, 248574664 ...
Before I can use these to play backgammon, you need to
reduce them to (derive from them) dice rolls. Perhaps
a very "complicated super-duper":) algorithm to accomplish
this would divide a given integer by 6, take the remainder
and add 1 to it.
I just type the above numbers without any special
consideration and I'm not going to real math, but if you
fed the above numbers to such an algorithm, it could
very well spit back:
.. 3, 3, 3, 3, 3, 3, 3 ...
Let's be a little more lenient and say that they produce:
.. 4, 6, 3, 4, 3, 6, 5 ...
Of course the algorithm may derive pairs of dice
numbers from each integer instead of one at a time, like:
.. 41, 62, 33, 41, 33, 62, 51 ...
A die has only six sides and when you reduce your unique
integers to dice roll, they are no longer unique but will
repeat millions of times.
But, that's ok. We are not worried about that. What we
are worried about is subsets of dice rolls repeating
too many times.
The subset "734, 156956, 2, 3456789, 888888, 1900, 248574664"
will never repeat in your set of 4 billion integers, simply
because even individual members of it never repeat.
But the subset "41, 62, 33, 41, 33, 62, 51" will surely
repeat, and it will repeat probably millions of times
in your main set of 4 billion rolls.
After converting your 4 billion integers to 4 billion
dice rolls, let's take the first 100 rolls in the set
and name it subset-A. If we want to know if subset-A
occurs in the main set more than once and how many
times more than once, what we would have to do is to
compare it to the subset of rolls 2-101, 3-102, 3-104,
4-105 and so on, 4 billion times. Since a given seed
and counter combination may point to roll number 1, or
2, or 3, or 19, or 1745, or 2534876, we would want to
also know if and how often the subsets of 100 rolls
starting at those numbers repeat in the main set. That's
4 billion comparisons, to be done 4 billion times.
If you wanted to eliminate some of those duplicatios,
then as a second step you would have to do billions of
more comparisons, so that you won't be undoing the very
same changes you have made. This is the gigantic task
I was talking about.
Since nothing is impossible, let's say somebody took
the time to go through the first stage of the above
process and counted the recurrence of each of the 4
billion subset of 100 rolls (ignoring reciprocal
duplicates), and came up with:
Subset 1-100 reoccurs X1 times
Subset 2-101 reoccurs X2 times
Subset 3-103 reoccurs X3 times
....
Subset 18673-18773 reoccurs X1863 times
Subset X+N-X+N+100 reoccurs XN times
At that point, you can add up all your X? counts and
divide by 4 billion, to declare that "on the average,
any given set of 100 dice rolls reoccurs Y times in
this main set of 4 billion rolls", and from this you
can derive Z% as the rate of reoccurrence of any such
subset in the main set.
Theorically, repeating subsets will be evenly
distributed throughout your main set. So, if you
cut your main set in half, (took the first 2 billion
and discarded the rest) and then repeated the above
process all over; you will come up with the same Z%
rate of reoccurrence. If you keep reducing the size
of your main set to 800 million, 10 million, 50
thousand, 10 thousand, etc. you will keep getting
the same Z%.
In your set of 4 billion, you can't guarantee that
the first 100 rolls won't repeat right after in the
second set of 100 rolls. Besides this, the player is
not obligated to play 1000 rolls in an unbroken sequence.
He is free to pick and choose seed/counter values manually.
He may choose 1765/42 for the first game, 198542/76453
for the second game and by golly they may happen to
be duplicate sets. So, if Z% repeat rate doesn't pose
a problem within a set of 4 billion rolls, neither
will it cause a problem within a set of 10 thousand
rolls.
If a player plays 2000 games in one sitting, he would
have gone through almost a quarter of a main set having
only 10000 rolls, and that may may pose a problem. But
this is unlikely and nor a player would have the memory
of an elephant to remember hundreds of sequences of
dozens of rolls. Then, I would say a set of a few thousand
rolls should be more than plenty enough, shouldn't it be?
I don't know if I have a language deficiency to make
myself understood, but I think this time I have tried
all I can to explain what I was trying to make a point
about. I hope you or others don't come back again with
how many billions of unique integers can be generated
with 16 bits, 32 bits, etc. That's irrelevant regarding
the issue I'm trying to raise...
>> Just let me type a 14 digit number and you
>> figure out at what position in your great circular
>> set of rolls you want to start rolling from. From that
>> point on, just roll me dice and when you reach
>> 100,000,000,000,000 wrap around and keep rolling...
>> Do you see why I'm dwelling on the uniqueness implied
>> by the "chunks" (i.e. "counters") of sequences...?
>No, I don't! Uniqueness of _subsequences_ isn't necessary or
>sufficient for the sequence as a whole to be `random'. Besides,
>if you look at small enough subsequences in a large enough
>sequence, then they won't be unique anyway (since there are a
>finite number of possible subsequences of a given length, and
>if you take a large enough sequence then eventually you'll
>exhaust them all).
Ok, what you are saying here is relevant to my point.
There are two questions that can be asked here: 1- What
is the limit of the maximum sequence that can be produced
with only pairs of numbers from 1 to 6, with unique subsets
of size N? 2- Whether a 5 digit subset (i.e. JF's counter)
should be considered "small enough" for a 9 digit main set
(i.e. JF's seed)?
I'm not all that good at doing actual computations and
don't know the answer to these questions. I can only
guess that the answer to the first one may be smaller
than one would expect. For the second one, I think
99999 is not small but rather large enough to not be
"expected" to repeat in a subset expressed by 14 digits.
If you take a subset made of 99999 times the same dice
combination, let's say 6-6, you can create another unique
subset by merely changing one roll to, let's say 2-1.
Then, by merely changing the location of that 2-1 by
one position each time, you can create 99999 more unique
subsets already. By doing the same all over using the
dice combination 3-1 instead of 2-1, you would double
that number, etc...
>> If an algorithm generating *non-random* sequences, has
>> a repeat-rate of X, that's not going to change or get
>> any better by using smaller or larger numbers. What may
>> be JF's repeat rate? And if that repeat rate is not going
>> to get any more visible by reducing the counter/seed
>> sizes to X digits, why not use smaller digits instead...?
>By `repeat rate', I assume you mean period. The period of
>the example generator I gave above is 2^32, and that's a...
No, this is not what I meant. I tried to explain above
what I meant by this (i.e. how many seed/counter
combinations is likely to produce the exact same
sequence of let's say 40 rolls of dice in a row).
>I have been trying to kill these threads for weeks now without
>any success :-)
Why? I think this is interesting stuf...
>I think we've digressed too far and lost the point of the
>original accusation. The claim is that Jellyfish `cheats' with
>some (unspecified) collusion between its dice generator and
>its choice of play (ie. evaluation function), correct?
It came to this based on the speculation that if the
dice rolls are pre-recorded and come from a file,
then it could be possible to manipulate them in some
ways to make them "lead" to certain positions.
>Well, I can prove that it _doesn't_ cheat with only two
>assumptions. If it does cheat, then one of my assumptions
>must be wrong. Please give me a counterexample demonstrating
>which one is unjustified. The assumtions are:
>1) The dice rolls depend ONLY on the seed/counter and NOTHING
> else (not the board position; not the match score; not the
> phase of the moon).
Ok, but if they come from a file, instead of from an
algorithm, then they could be manupulated. Impression
of an implied uniqueness among counters (subsets) leads
one to the argument that this would be impossible (or at
least impractical) through an algorithm and to the idea
that they may be being fetches from a pre-recorded file
instead. If no such uniquesness implied, the only
remaining question is "why such large numbers"? (which
in itself is irrelevant to the subject of "cheating" as
you noted above).
MK
Murat Kalinyaprak
>mu...@cyberport.net (Murat Kalinyaprak) writes:
>Well, I guess that the people offering to back JF at $10/point would
>be willing to back it for any amount you choose. There is, as far as
>I know, at least one who is would be willing to do that. In previous
>posts, he offered to pay for flight and hotel, only the person would
>play at least a certain amount of games, the number of which I don't
>remember. So, to me it's more the other way around. It's you who are
>claiming that Jellyfish cheats and you can beat it with manual dice.
>They don't know you but they know Jellyfish, and therefore they have
>generously offered you to play for any amount you want.
And I'm still trying to figure out why any assumedly
sane/reasonable person would do such a thing? I didn't
come here saying that I had never heard of Jellyfish
but that I could beat it no matter what it was, how
good it was. I had already played a good number of
games/matches against it haphazardly, and I knew that I
was beating it more often than it was beating me. But
still, I didn't come here to solicit bets. As the
discussions progressed in a certain direction, I decided
to play some sets of matches with clearly separated
beginnings and ends (apart from experimenting with it)
and reported the results here.
My attitude was never to brag about myself. To the
contrary, I was saying: "Hey guys, look, I'm just a
nobody/amateur but I can beat your much praised
Jellyfish". It was either that JF was as bad as me,
or that I was as good as Jellyfish. Since I didn't
make much of myself, my inclination was to put down
JF to my level and when I did that people came to its
defence (which included offering me bets, etc.)
If you had told me that you had/were beating JF, I
would have taken your word for it and never do as a
stupid thing as betting money on JF against you. So,
I don't understand why some people would bet money
on JF against me when I say that I'm beating it. As
far I as can guess, there may be a few of reasons
for it.
1- They don't believe what I'm saying. If this is the
case, all I have to say is that I don't need to take
on some people's bets just to prove my honesty. (I had
already expressed earlier that I have no intentions to
make a career out of gambling either). As far as this
subject goes, there is not much more to say.
2- They think my winning 3 out 3 matches with manual
dice and 6 out of 10 matches with automatic dice roll
was just an unusual luck, and that once I run out of
it, JF will beat me forever after. This may very vell
be the case but it reduces backgammon down to luck a
little too far and makes one wonder why even bother
with moving pieces around on a board? Why don't they
propose instead that we simply roll dice for money?
3- Other reasons that elude me...???
>Moreover, as
>for not knowing anything about you, you have supplied proofs to this
>newsgroup about your knowledge of backgammon. I haven't analyzed the
>matches yet, but judging from some positions from these and comments
>that were made, I'd say that they are not likely to lose money.
If you haven't even analyzed the few games I posted,
what "proofs I supplied" that you are talking about?
I have already admitted that I don't quite know how
the use the cube efficiently/properly, I have no clue
about pips/equities/etc., since I haven't played the
game much in this country I don't know the "lingo",
and all the of such things "non-essential" to the game.
I played the game for about 30 years and have an idea
about how it's played. With a glance at the board, I
can tell whether I'm ahead or behind, by how much, etc.
sufficiently good enough for the purposes I played the
game for so far in my life. Frankly, at the expense of
offending some people, calculating odds/equities/etc.
to the degree I see being done here, a little too
"nerdish". Even trying to slow down myself playing
against JF, I can't help making most of my move as fast
as I can glance at the board and click the mouse button
(a la Mediterranean/Middle-Eastern coffee house style).
It would be more fun for me to play backgammon using
a chess-clock instead of using a doubling cube. Now,
don't take these wrongly again as my bragging about
myself. It's the other way around. I'm exposing my
deficiencies against a computer based player. Yet,
I'm still beating Jellyfish at level 7/1000 more often
than not, with/without the cube, with/without manual
dice rolls. If this rubs off some people the wrong
way, I can't help it. And there is nothing they can
do about it either (beyond offering bets, that is)...
>Take
>this comment as you want, it is not my wish to ridicule your ability
>to play backgammon.
You guys can take my comments as you wish also. I'm
not making comments about or trying to ridicule a
computer game either. For example, I have never played
against the one called Snowie and never made a comment
about it, other than its price is too high for any
computer game program in my opinion. I don't know
whether I can beat Snowie or not, and I'm not talking
about what I don't know. As far as about JF, from where
I'm sitting, it all boils down to this: "Either I'm an
undiscovered backgammon-champ-to-be, or JF is not what
it's cracked up to be". I have no problem letting you
choose which it is...
MK
Gavin Anderson
John Goodwin wrote in message <35b6d254....@news.demon.co.uk>...
(snip)
>>>The man agrees, and asks what stakes the master would like to
>>>play for.
>>>
>>>"Any amount you choose", replies the master.
>>>
>>>"WHAT?!? Do you have any idea who I am?"
>
>>>"No, none at all. And that's why I'll let you choose the stakes."
>
>
>I was hoping someone else would ask, but as they haven't I'll have to
>admit my stupidity and do so.
>
>What does this tale mean?
>
>I would normally expect it to be followed by the grand master getting
>beaten, and a point being made about taking your superiority for
>granted.
>
>As it is, it doesn't make much sense. The grand master must be fairly
>stupid if he makes wagers *because* he has less information than he
>might have.
>
>OK, I've got a damp cloth ready to wipe the egg from my face.
>
>JG
The way I see it - the point is that the grand master knows he's world
class, and he knows the name/reputation of any opponent worthy of him
(rumour, legend, experience etc). So anyone he's never heard of must
therefore not be a world-class player, and so he can take any stakes and
win. Anyone who was world-class WOULD de facto be known about.
So when Kit Woolsey meets me on the train, he can probably except any stakes
with confidence - because if I WAS world class my name would be Nack Ballard
(or something). (And if I was a decent cheater he would have seen me at
Monte Carlo trying it on).
The story was presented as an anecdotal reply to the statement 'how do you
know how good I am - you don't know me' to show why it might be valid for
someone to say 'I know you're not a world-class player, because I've never
heard of you'.
Gavin Anderson
Claes Thornberg
So, if I understand you correctly, what YOU are worried about is that
a subset of the sequence of rolls MAY occur more than expected. Well,
that hypothesis is easy to test, though it may take some time. If you
really want to test this, I think you have to test if any sequence of
length N (N = 1, 2, ...) occurs more than expected. But you're better
off reading books on statistical analysis than trusting my advice.
The probability of the sequence "41, 62, 33, 41, 33, 62, 51" is quite
easy to calculate. It's 1/18^5 x 1/36^2 = 1/2448880128. So this means
that the above sequence should occur less than two times. Not, as you
suggest, millions of times. I don't know what a significant deviation
from the expected is, but if that sequence occurred more than, say 10
times, I would worry about the randomness of a roll generator.
> After converting your 4 billion integers to 4 billion
> dice rolls, let's take the first 100 rolls in the set
> and name it subset-A. If we want to know if subset-A
> occurs in the main set more than once and how many
> times more than once, what we would have to do is to
> compare it to the subset of rolls 2-101, 3-102, 3-104,
> 4-105 and so on, 4 billion times. Since a given seed
> and counter combination may point to roll number 1, or
> 2, or 3, or 19, or 1745, or 2534876, we would want to
> also know if and how often the subsets of 100 rolls
> starting at those numbers repeat in the main set. That's
> 4 billion comparisons, to be done 4 billion times.
> above
See my comment above. You must test that the occurrence of a sequence
does not deviate too much from the expected, no matter what sequence,
or how long it is. I suggest you start doing that now, and keep us up
to date with your progress. If you provide data in convienent format,
I think others might be willing to do the analysis and tell us if you
are right in your suspicions or not.
By the way, did you know that the probability of a sequence of length
100 is about 1/10^150. So if you find a sequence that occurs twice, I
would be surprised, since most sequences of length 100 don't occur at
all.
> [SNIP]
>
> >No, I don't! Uniqueness of _subsequences_ isn't necessary or
> >sufficient for the sequence as a whole to be `random'. Besides,
> >if you look at small enough subsequences in a large enough
> >sequence, then they won't be unique anyway (since there are a
> >finite number of possible subsequences of a given length, and
> >if you take a large enough sequence then eventually you'll
> >exhaust them all).
>
> Ok, what you are saying here is relevant to my point.
> There are two questions that can be asked here: 1- What
> is the limit of the maximum sequence that can be produced
> with only pairs of numbers from 1 to 6, with unique subsets
> of size N?
This question should be easy to answer, at least partially. There are
21 unique rolls of the dice, regarding 63 and 36 as the same roll. It
should therefore be possible to make a sequence of 21 rolls where the
subsequences of length one occurs exactly once. It's easy to see that
this is indeed the case. By the way, there are 21! such sequences. It
is easy to see that the number of unique two roll sequences is 21x21.
If we make a sequence, which contains each of the 21x21 subsequences,
it will have length 21x21, the number of included subsequences, times
2, the length of each subsequence.
If N is the length of subsequences, we conclude that you can't make a
sequence with more than 21^N subsequences of length N, without having
to repeat any subsequence of length N. However, sequences must repeat
themselves, at least to some degree. In a sequence of 36 rolls, you'd
expect every doublet to occur once, and any other roll exactly twice.
And in a sequence of 1296 rolls, every doublet should occur 36 times,
every other roll 72 times, all two roll subsequences without doublets
should occur 4 times, all two roll subsequences with just one doublet
should occur twice, and any subsequence with two doublets, of course,
should occur exactly once. But, it's easier to regard every roll as a
unique one, that is, 21 and 12 are different rolls. Then there are 36
different rolls. Therefore the number of unique sequences of length N
is 36^N. The length of a sequence, which contains every single one of
these subsequences, is 36^N, the number of subsequences, times N, the
length of the subsequences. In other words, for a sequence to contain
all 36^N unique subsequences of length N, it must be Nx36^N rolls. If
shorter, it can not contain every unique subsequence of length N, and
if it's longer it must contain some sequence more than once.
However, making a sequence of Nx36^N rolls does not guarantee that it
is a sequence where every subsequence occur exactly once. Nor does it
mean that any sequence of Nx36^N rolls, where each unique subsequence
of length N occurs exactly once, is random.
When reading through my comments, I saw that I may have misunderstood
your question. I don't think so, but it wouldn't surprise me. You may
want to how many different sequences you can make, where any sequence
contain all unique subsequences of length N. Remember that sequences,
to some degree, must repeat themselves due to doublets only occurring
half as often as non-doublets. In my comments above, there is a hint.
There are N! unique permutations for any set of N unique items. Since
there are 36^N unique subsequences of length N, there must be (36^N)!
permutations of these subsequences.
> 2- Whether a 5 digit subset (i.e. JF's counter)
> should be considered "small enough" for a 9 digit main set
> (i.e. JF's seed)?
>
I admit that I don't understand this question at all. It might be I'm
stupid, but I hope that's not the case.
>
[SNIP]
>
> >By `repeat rate', I assume you mean period. The period of
> >the example generator I gave above is 2^32, and that's a...
>
> No, this is not what I meant. I tried to explain above
> what I meant by this (i.e. how many seed/counter
> combinations is likely to produce the exact same
> sequence of let's say 40 rolls of dice in a row).
I would say it's unlikely that anyone, except by reverse engineering,
will find any such pair of counter/seed combination. And to be frank,
I don't believe that there is such a counter/seed combination.
>
[SNIP]
>
> It came to this based on the speculation that if the
> dice rolls are pre-recorded and come from a file,
> then it could be possible to manipulate them in some
> ways to make them "lead" to certain positions.
>
This sounds pretty vague. Could you please explain what you mean? Are
you suggesting that there is some sequence of rolls which repeats? Or
that that some sequence occurs more often than others? Or that Jelly,
when seeing that things are going certain ways, changes the sequence,
from which it selects the rolls? Or that it's playing a position with
some kind of foreknowledge of what rolls are coming? Or that there is
some other way of "cheating" which goes undetected, even though there
are several thousands of players using the software? Or that you have
actually detected the "cheating" but you can not form a hypothesis of
how it is done, and therefore we cannot test whether you are right or
not? Or that Jellyfish always plays suboptimal moves, and these moves
are good when playing with manual dice, but they become optimal using
the built-in dice generator? As you see, I'm interested in seeing you
elaborating on how this manipulating is done. Please, keep this topic
alive.
>
> >Well, I can prove that it _doesn't_ cheat with only two
> >assumptions. If it does cheat, then one of my assumptions
> >must be wrong. Please give me a counterexample demonstrating
> >which one is unjustified. The assumtions are:
>
> >1) The dice rolls depend ONLY on the seed/counter and NOTHING
> > else (not the board position; not the match score; not the
> > phase of the moon).
>
> Ok, but if they come from a file, instead of from an
> algorithm, then they could be manupulated. Impression
> of an implied uniqueness among counters (subsets) leads
> one to the argument that this would be impossible (or at
> least impractical) through an algorithm and to the idea
> that they may be being fetches from a pre-recorded file
> instead. If no such uniquesness implied, the only
> remaining question is "why such large numbers"? (which
> in itself is irrelevant to the subject of "cheating" as
> you noted above).
I would also like you to comment on the difference between "cheating"
and "manipulating". You say earlier that Jellyfish might "manipulate"
the dice or change its way of playing because of foreknowledge of the
rolls. I would say this is cheating, but obviously you don't agree.
Gary Wong
backgammon-playing program; this article is in response to queries
about whether its pseudo-random dice rolls are (or can be) `fair'.)
I think these posts are getting far too long, so rather than quoting your
entire article I understand you still seem to have four main concerns:
* Is it possible to produce a good random number generator at all?
The answer to this question is an emphatic `yes'. The best generator I
am aware of (I believe it holds the world record for longest period
and maximal k-distribution) is the Mersenne Twister [1]. Its period
is 2^19937-1 and it produces 32-bit integers -- that means you can
produce over 10^6000 integers before the sequence repeats! If you
thought the range of Jellyfish's seed and counter were overly large,
then this is just too astronomical to think about. I'm at a loss to
think of an analogy to how big this quantity really is. If you wrote
out the number, it would take you about 2 pages to do it. If every
particle in the universe contained a universe of its own, and you
wrote one random number on each particles within, you'd run out of
particles long before you even scratched the surface of the numbers this
algorithm can generate. As long as games don't contain repeated positions,
then you could play every possible game of backgammon with these dice
before the sequence repeated -- not just every single possible
sequence of rolls, but every possible way of _playing_ those rolls.
And still not scratch the surface :-)
Not only is the period extraordinarily good, but the distribution is
excellent too. It is proven to be 623-distributed to 32-bit accuracy
(which means that EVERY subsequence of 32-bit integers of up to length
623 is represented exactly equally). In smaller ranges the length
increases -- one roll of a pair of dice (ie. 36 combinations) contains
about 5.17 bits of information; so 6 bits suffice for one roll. At
6-bit accuracy, this generator is 3115-distributed -- this means that
EVERY possible list of 3,115 consecutive dice rolls (and there are
36^3115 of them!!) is represented within this generator's sequence,
and furthermore they are distributed equally. There's nothing
probabilistic about this measurement, it is PERFECT. Earlier you
mentioned you would `bow in deep reverence' to an algorithm posted in
a newsgroup that could generate 1000 unique sequences of 100 rolls
-- well, there's not much point posting it here, but it's freely
available on the web page referenced above. And it generates 36^3115
unique sequences of 3115 rolls -- please bow in deep reverence to
Matsumoto and Nishimura :-)
* Does a generator with good properties in the range (for example) 0..4
billion necessarily imply that it's also a good generator in the range
1..6 (for backgammon)?
Yes. A perfect random number in the range 0..4 billion contains 32
bits of `entropy' (not a rigorous use of entropy, since the numbers
are only psuedo-random; but the distinction is unimportant); any
function that equally distributes integers from this domain to the
range 1..6 (eg. the `modulo 6' example you gave) will preserve about
2.58 bits of entropy (ie. log base 2 of 6). A less-than-perfect
generator would not produce 32 bits of entropy, but in practice as
long as it produces well over 2.58 bits, and every bit is used in the
mapping function (ie. no information is discarded), then the
distribution over the range 1..6 will be close to perfect. A
rigourous proof requires some information theory -- if you are
interested in verifying the result yourself I suggest you look for
material on information theory on the WWW, eg. [2].
* Is having (some large number of) seed/counter values justified?
If the internal state of the random number generate contains as much
information (ie. as many bits) as the seed and counter values combined,
then yes, it definitely is. The generator _requires_ that much information
to be capable of generating its full range of output. If Jellyfish's
generator's state is at least 32 bits (and it almost certainly is), then
lower bounds on the seed and counter would be artificially limiting its
range. Why would you want to do that?
You shouldn't assume that the fact that the state information is
divided into two parts is a `promise' that every subsequence with
a distinct seed but equal counter is unique. They almost certainly
_are_, but that's beside the point. The combined value _needs_ to be
that large to allow you access to the entire `sequence space'. Michael
Zehr points out that having two values (seed and counter) instead of
one allows you to compare rollouts of the same position with the `same'
dice (if each rollout starts at seed n, n+1, etc. and counter 0, and
each roll continues with the next counter, then multiple rollouts of
the same position played different ways will be measured with the
same dice). I know I claimed yesterday that duplicate dice don't help
over the course of a match -- there's some justification for it here
but I've digressed far too much already :-)
There is an infinitesimal possibility that perhaps (for example) the
subsequence of 100 dice rolls from seed 314159, counter 0 happens to be
the same as the subsequence of 100 dice rolls from seed 271828, counter
0. If this were true, it would be no more significant than a slight
curiosity and a coincidence in number theory (just like finding that
371 is the sum of the cube of its digits, 3^3 + 7^3 + 1^3). There
is NO way (with any reasonable generator) that large portions of the
sequence space are duplicated like this (eg. seed 0 is identical to
seed 65536, and seed n is identical to seed 65536+n, etc.) -- that would
be equivalent to showing that the generator's period was smaller than
its sequence space, and there are so many generators published that
have `full' periods that it's trivial to find one.
* Given that it is possible to produce a good generator, is Jellyfish's
generator adequate?
Without knowing the details of Jellyfish's generator it's impossible to
say for sure, but personally I have no doubt that it is. Basic linear
congruential generators are adequate and so ubitiquous in computer science
it is inconceivable that the author of Jellyfish was unable to code
at least a reasonable LCG. It is a very safe assumption that Jellyfish's
generator lies (in quality) somewhere between a basic LCG (adequate) and
the Mersenne Twister referred to above (astronomically good).
I don't have the time, inclination or ability to discuss random number
generators any further, but there is a HUGE amount of knowledge published
on the subject. Linear congruential generators are over 40 years old
and have been studied inside out. If you are interested, look at [3] for
all kinds of useful references.
Follow-ups to sci.math.num-analysis.
Cheers,
Gary.
[1] http://www.math.keio.ac.jp/~matumoto/emt.html
[2] http://www.math.washington.edu/~hillman/Entropy/infcode.html
[3] http://random.mat.sbg.ac.at/links/
Murat Kalinyaprak
> Murat Kalinyaprak wrote:
>>I'll post the results along with some statistics
>>and comments.
>>Would anyone care to comment on what kinds of
>>statistics would be meaningful/beneficial...?
> Beneficial for me would be things like:
> Max/mean average equity difference between your moves and JF's
> top plays, per game.
> Proportion of your moves, per game, that JF disagrees with by
> more than 0.03 equity.
I wish I could give you such stats, but I'm only using the
trial (player) version of JF which doesn't have such
functions and I have no clue as to how to calculate equity,
etc. manually.
> Breakdown of types of game: how often you played a backgame,
> defended against a backgame, played a mutual holding game,
> racing game, etc. Plus the above two measures, and win rates,
> split by game type, if that's possible.
I wish I had done this as I was playing the games. To do it
now, I would have to reload/replay each game, which would
take some time. If anybody is interested in analyzing the
games to do this or any other kind of stats, I'll be happy
to make them available.
> As you seem to play a Middle Eastern style game from your
> posts (I haven't looked at the games you've posted), this
> could be interesting.
On top of this, I had also used the term "coofee house"
style, which means a pretty fast paced play with lots
of talking/riddles/jokes...
Your having paid attention to my previous comment on this
prompted me to take some "Middle Eastern stats":), i.e.
those 100 games were played over 3 days (in 4-5 runs of
2-3 hours each), taking about 11-11.5 hours in total (I
derived this from date/time each ".gam" file was saved).
This works out to slightly less than 7 minutes per game.
Ending counter value indicates a total of 5053 moves and
that works out to about 8 seconds per move.
I don't know if this is average, twice longer or four
times shorter than a typical pace expected, but I just
I'd give you some stats for the fun of it (and to make
up for the ones you asked for but I couldn't provide).
MK
Murat Kalinyaprak
> John Goodwin wrote:
>>I would normally expect it to be followed by the grand master getting
>>beaten, and a point being made about taking your superiority for
>>granted.
>>As it is, it doesn't make much sense. The grand master must be fairly
>>stupid if he makes wagers *because* he has less information than he
>>might have.
>>OK, I've got a damp cloth ready to wipe the egg from my face.
> The way I see it - the point is that the grand master knows he's world
> class, and he knows the name/reputation of any opponent worthy of him
> (rumour, legend, experience etc). So anyone he's never heard of must
> therefore not be a world-class player, and so he can take any stakes and
> win. Anyone who was world-class WOULD de facto be known about.
I agree with John Goodwin here, because the story doesn't
make sense unless you add some assumptions to it, like you
are doing here. The problem with your above reasoning is
that the master never asked for the other guys name.
> So when Kit Woolsey meets me on the train, he can probably except any
> stakes with confidence - because if I WAS world class my name would be
> Nack Ballard (or something).
I gather these are names of world-class players. Now, imagine
that Kit W. runs into Nack B. on that train. Kit had heard of
Nack but had never met him or seen his pictures, etc. Unless
he asked what the other's name was, Kit wouldn't know that he
had met Nack.
Also, the offer was not to play 100 chess games of 1 point,
but just one game. And backgammon ain't quite chess.
So, Kit lets Nack choose the stakes and Nack says "Whatever
you got in your pocket". They play a game, Nack gets a few
nice rolls and Kit is wiped out in a couple of minutes...
This is about the only "lesson" I myself can derive from the
above incomplete story.
> The story was presented as an anecdotal reply to the statement 'how
> do you know how good I am - you don't know me' to show why it might
> be valid for someone to say 'I know you're not a world-class player,
> because I've never heard of you'.
The problem is, how do you know if you have ever heard of
someone or not, if you never even bother to ask his/her name?
In this case, the writer who posted the story knew my name
(which is in the article headers), but the anectode still
doesn't apply because the story is missing a necessary
element...
I was going to say that to be a good backgammon player,
one needs to be a "through thinker", but I'm not going
to say it... :)
MK
Rodrigo Andrade
2- This may very vell
be the case but it reduces backgammon down to luck a
little too far and makes one wonder why even bother
with moving pieces around on a board? Why don't they
propose instead that we simply roll dice for money?
No matter what you do, no matter what you say, backgammon is still a game that involves chance. Not like slots, but there's still a luck factor. Just because you win (from your own data) 100% of the games when you roll the dice, and only 60% when JF does, it doesn't necessarily mean it's cheating! Maybe you're just running out of luck! Believe me, it took me 1 year to understand that. I've been playing casual chess forever, and it's an entirely different thing. You outplay your opponent, you win. This is not always the case w/ backgammon. In many occasions, I outplay my opponent and still lose!
I bet that you sometimes lose to JF even when you're rolling the dice yourself. How come?? Is it cheating?? NO!! That's just luck.
RODRIGO
Murat Kalinyaprak
> 2- This may very vell
> be the case but it reduces backgammon down to luck a
> little too far and makes one wonder why even bother
> with moving pieces around on a board? Why don't they
> propose instead that we simply roll dice for money?
>No matter what you do, no matter what you say, backgammon
>is still a game that involves chance. Not like slots, but
>there's still a luck factor.
That's what I mean also. You may see me even emphasize
the luck factor, but not reduce it to luck too much.
This newsgroup is a whole different world of backgammon
for me. The way I played this game before, if someone
beat me 3 matches in a row (i.e. 5 point matches that
last at least 3 games each without cube and back-gammons),
I would shake his hand and say he is a better player. I
could say he got lucky for a few games, a match or two,
but if I still said that he had just gotten lucky after
loosing 3 matches in a row, I would be looked upon as a
sour looser and accused of reducing the game too far down
to luck. And if I asked for another match right after
loosing three in a row, I would risk hearing comments
like "I know loosers never get enough of loosing, but
I'm not your backgammon teacher. Go learn how to play
first, then come back and we'll play more" or something
similar. It's just known through "collective experience
and knowledge" that I would be very unlikely to beat
that person just 10 minutes or 2 hours later. I had never
heard of having to play 100, 200, 3000 games in order to
eliminate the luck factor and figure out how one measures
up against an opponent. I personally believe that the luck
factor is eliminated rather quickly between uneven players
and may take longer between closely matched players, but
after a certain point this really becomes a matter of
breaking the tie, more than eliminating the luck factor.
I can even understand 100, 200 games, but 3000 games just
to prove that one can beat a computer game...? Is it even
worth that much to know...?
>Just because you win (from your own data) 100% of the games
>when you roll the dice, and only 60% when JF does, it doesn't
>necessarily mean it's cheating! Maybe you're just running out
>of luck!
Actually, after the 3 matches I reported here, I haven't
played much rolling manual dice. I figured that I would
grab the bull by the horns (i.e. let JF roll the dice)
and see what happens. Now, I even wonder if it wouldn't be
easier to beat JF with automatic dice? (related to some
speculations I had made earlier)
MK
Patti Beadles
pretty much know each other.
I'm not a top player. But if a random stranger approached me right
now and offered to play me for any stakes that I could reasonably
afford, I'd agree without hesitation. The likelihood of that person
being significantly better than I am is relatively small.
-Patti
--
Patti Beadles |
pat...@netcom.com/pat...@gammon.com |
http://www.gammon.com/ | The deep end isn't a place
or just yell, "Hey, Patti!" | for dipping a toe.
Bill Triplett
any game, the longer two opponents play each other, the more they adapt to
each others game. I played 1000 pt match against JF. I won 442 pts to its
558. When I play people however, I do not do so well. After playing
different people for awhile I can go back to JF and get beat most of the
time until I've played it for awhile. Then I start winning more.
My brother and I used to play tournament chess. I consistently beat him,
and consistently beat people that beat him. Then we played a 'match' one
summer and we tied 272 games apiece with 11 draws. The number would suggest
we were equals in skill. However, I continued to beat those that beat him.
After we quit playing each other so frequently, we are now back to the same
thing as before. I consistently win. If however, we play a long enough
match, we would probably tie again.
BT
--
Bill Triplett
Murat Kalinyaprak wrote in message <6pcbf1$mld$3...@news.chatlink.com>...
>In <6pbq60$na8$1...@news4.wt.net> Rodrigo Andrade wrote:
>
>> 2- This may very vell
>> be the case but it reduces backgammon down to luck a
>> little too far and makes one wonder why even bother
>> with moving pieces around on a board? Why don't they
>> propose instead that we simply roll dice for money?
>
my_...@hotmail.com
mu...@cyberport.net (Murat Kalinyaprak) wrote:
> The way I played this game before, if someone
> beat me 3 matches in a row (i.e. 5 point matches that
> last at least 3 games each without cube and back-gammons),
> I would shake his hand and say he is a better player. I
> could say he got lucky for a few games, a match or two,
> but if I still said that he had just gotten lucky after
> loosing 3 matches in a row, I would be looked upon as a
> sour looser and accused of reducing the game too far down
> to luck.
Think of it this way. Suppose you flip a coin 3 times,
and it comes up heads each time. Is that so unreasonable?
Now, I'm not equating a backgammon match to flipping a coin.
But suppose you lost three matches in a row, and you were in
fact of exactly equal skill to your opponent. The chances
of you losing those 3 games, as with flipping "heads" three
times in a row, are 1 in 8. So that's not yet proof that he
is truly better than you.
> I can even understand 100, 200 games, but 3000 games just
> to prove that one can beat a computer game...? Is it even
> worth that much to know...?
>
It makes no difference to me to know who's better than whom
(only who's better than ME!)
But one could argue that after 3000 games, being that JF is
such a good tutor, you could indeed improve your game enough
to be as good or better than it.
-----== Posted via Deja News, The Leader in Internet Discussion ==-----
http://www.dejanews.com/rg_mkgrp.xp Create Your Own Free Member Forum
John Greenwood
>But one could argue that after 3000 games, being that JF is
>such a good tutor, you could indeed improve your game enough
>to be as good or better than it.
It may teach you all you (one) know(s), but that is not the same as
teaching you all it knows:)
---
John Greenwood
Murat Kalinyaprak
>mu...@cyberport.net (Murat Kalinyaprak) writes:
>> But the subset "41, 62, 33, 41, 33, 62, 51" will surely
>> repeat, and it will repeat probably millions of times
>> in your main set of 4 billion rolls.
>So, if I understand you correctly, what YOU are worried about is that
>a subset of the sequence of rolls MAY occur more than expected.
Yes, this is definitely one thing I'm concerned about.
>that hypothesis is easy to test, though it may take some time. If you
>really want to test this, I think you have to test if any sequence of
>length N (N = 1, 2, ...) occurs more than expected. But you're better
>off reading books on statistical analysis than trusting my advice.
I'm not sure if reading books would help, because I'm
not corcerned about just any situation but mainly in
the case of JF (at least as far as these threads go).
>The probability of the sequence "41, 62, 33, 41, 33, 62, 51" is quite
>easy to calculate. It's 1/18^5 x 1/36^2 = 1/2448880128. So this means
>that the above sequence should occur less than two times. Not, as you
>suggest, millions of times.
Ok, I was exaggerating. I somehow feel that it would
occur more than two times though. But I'm not all that
good with calculating such things and if nobody else
objects to your above numbers, I won't mind believing it.
>I don't know what a significant deviation
>from the expected is, but if that sequence occurred more than, say 10
>times, I would worry about the randomness of a roll generator.
Yes, and this would apply to shorter sequences of only
2, 3, 4 rolls also. Is there such thing as "inherently"
hard to play sequences of X rolls regardless of board
positions? If so, then any such high rates of occurrences
may be evaluated from a certain perspective...
>See my comment above. You must test that the occurrence of a sequence
>does not deviate too much from the expected, no matter what sequence,
>or how long it is. I suggest you start doing that now, and keep us up
>to date with your progress. If you provide data in convienent format,
>I think others might be willing to do the analysis and tell us if you
>are right in your suspicions or not.
If I'm provided the data in a convenient format, I can
do some of the analisys myself also. I'm interested
particularly in JF's dice roll sequences. As it's now,
the fastest way to accumulate data is to keep clicking
on the "running-man" icon to have JF auto-play games and
then look at the log files it generates. That would take
years to do. But if the JF dice rolls were available in
chunks manageable on an average PC, let's say 10000 seeds
at a time recorded in a simple file - 1 byte per roll,
then it would be practically feasible to do such analyses
and I sure wouldn't mind doing/seein them. If there were
nothing unusual about them, I would be one step closer
to become another JF admirer/advocate/defender/etc...
>If N is the length of subsequences, we conclude that you can't make a
>sequence with more than 21^N subsequences of length N, without having
>to repeat any subsequence of length N. However, sequences must repeat
Not to quote this whole section, I think this is the
number I was looking for. It's finite but obviously
it's much larger than JF's 999999999*99999, even when
N is much smaller than 40 (which was the smaller number
number of rolls per game, that I had proposed).
>> 2- Whether a 5 digit subset (i.e. JF's counter)
>> should be considered "small enough" for a 9 digit main set
>> (i.e. JF's seed)?
>I admit that I don't understand this question at all. It might be I'm
>stupid, but I hope that's not the case.
Actually the "9" above should be "14". The question
was whether a subsequence of 99999 should be considered
"likely" to occur more than once within 99999999999999.
>> No, this is not what I meant. I tried to explain above
>> what I meant by this (i.e. how many seed/counter
>> combinations is likely to produce the exact same
>> sequence of let's say 40 rolls of dice in a row).
>I would say it's unlikely that anyone, except by reverse engineering,
>will find any such pair of counter/seed combination. And to be frank,
>I don't believe that there is such a counter/seed combination.
Then it would be even less likely to occur in a
seed/counter combinations limited to 99999 or 999999
rolls. That's why my original question had asked why
bother with larger numbers (which is mostly irrelevant
by now anyway).
>> It came to this based on the speculation that if the
>> dice rolls are pre-recorded and come from a file,
>> then it could be possible to manipulate them in some
>> ways to make them "lead" to certain positions.
>This sounds pretty vague. Could you please explain what you mean? Are
>you suggesting that there is some sequence of rolls which repeats? Or
>that that some sequence occurs more often than others? Or that Jelly,
>when seeing that things are going certain ways, changes the sequence,
>from which it selects the rolls? Or that it's playing a position with
>some kind of foreknowledge of what rolls are coming? Or that there is
>some other way of "cheating" which goes undetected, even though there
>are several thousands of players using the software? Or that you have
>actually detected the "cheating" but you can not form a hypothesis of
>how it is done, and therefore we cannot test whether you are right or
>not? Or that Jellyfish always plays suboptimal moves, and these moves
>are good when playing with manual dice, but they become optimal using
>the built-in dice generator? As you see, I'm interested in seeing you
>elaborating on how this manipulating is done. Please, keep this topic
>alive.
The key to whether I can elaborate on these is to
first know whether some sequences repeat more than
what would be considered "normal/expected". I already
gave some hints as to how/why this may be significant.
If I figured this one out, I may become interested in
developing yet another hard-to-beat computer backgammon
program myself... :)
>I would also like you to comment on the difference between "cheating"
>and "manipulating". You say earlier that Jellyfish might "manipulate"
>the dice or change its way of playing because of foreknowledge of the
>rolls. I would say this is cheating, but obviously you don't agree.
Well, if I could analyse millions of games and find
ways to make things particularly hard on you (vs. my
program), this is not quite "cheating" ouright (i.e.
spontaneously done as the situation requires, like
rolling what you need, etc.) because it's done ahead
of time and the player still has a "fair?" chance to
overcome such difficult positions if he's "really good".
Now, such things may be possible and still not apply
to JF. I don't think I'll be able to demonstrate anything
anytime soon, but I think I'll succeed in at least
making you guys wonder whether I'm not onto something
for real. The alternative being, my having to accept
the incredible reality that I may have been indeed a
world-class player all along, without knowing it... :)
MK
Claes Thornberg
(Murat Kalinyaprak) writes:
> In <yvk3ebr...@cuchulain.it.kth.se> Claes Thornberg wrote:
>
> >The probability of the sequence "41, 62, 33, 41, 33, 62, 51" is quite
> >easy to calculate. It's 1/18^5 x 1/36^2 = 1/2448880128. So this means
> >that the above sequence should occur less than two times. Not, as you
> >suggest, millions of times.
>
> Ok, I was exaggerating. I somehow feel that it would
> occur more than two times though. But I'm not all that
> good with calculating such things and if nobody else
> objects to your above numbers, I won't mind believing it.
The problem with feeling is that we are easily biased in our
judgements based on freak sequences, at least I am. But it's
a long way from feeling to certainty. For example, there are
a lot of people feeling that the FIBS dice are non-random. I
have, however, never seen any proof of them being rigged. On
the contrary, several convincing posts have appeared showing
that it's very likely that the dice really are random.
> >I don't know what a significant deviation
> >from the expected is, but if that sequence occurred more than, say 10
> >times, I would worry about the randomness of a roll generator.
>
> Yes, and this would apply to shorter sequences of only
> 2, 3, 4 rolls also. Is there such thing as "inherently"
> hard to play sequences of X rolls regardless of board
> positions? If so, then any such high rates of occurrences
> may be evaluated from a certain perspective...
The shorter a sequence is, the more it will occur. Therefore
the deviation from the expected will be larger. This applies
only to absolute figures. In relative terms it should be the
same. At least that's what I think. Could some correct me if
I'm wrong, here or elsewhere.
> If I'm provided the data in a convenient format, I can
> do some of the analisys myself also. I'm interested
> particularly in JF's dice roll sequences. As it's now,
> the fastest way to accumulate data is to keep clicking
> on the "running-man" icon to have JF auto-play games and
> then look at the log files it generates. That would take
> years to do. But if the JF dice rolls were available in
> chunks manageable on an average PC, let's say 10000 seeds
> at a time recorded in a simple file - 1 byte per roll,
> then it would be practically feasible to do such analyses
> and I sure wouldn't mind doing/seein them. If there were
> nothing unusual about them, I would be one step closer
> to become another JF admirer/advocate/defender/etc...
Not only is this method cumbersome and time consuming, it is
also error prone. We will miss opening doublets, since Jelly
will reroll such rolls without us ever noticing. It would be
much better if Fredrik, for instance, could produce a set of
rolls, which were sufficiently large, for anyone to analyze.
But I'm not sure if it's worth the effort to do this just to
gain another JF admirer/advocate/defender.
> >If N is the length of subsequences, we conclude that you can't make a
> >sequence with more than 21^N subsequences of length N, without having
> >to repeat any subsequence of length N. However, sequences must repeat
>
> Not to quote this whole section, I think this is the
> number I was looking for. It's finite but obviously
> it's much larger than JF's 999999999*99999, even when
> N is much smaller than 40 (which was the smaller number
> number of rolls per game, that I had proposed).
It depends on which N we choose. However, the number 36^N is
more correct, with subsequences being N rolls long. Choosing
N = 8.4, gives a number which is almost 10^14. But I believe
the possible values we can choose for seed and counter, have
little bearing on the quality of the generator. I think it's
added just for two reasons. First, it's important to be able
to do rollouts with the same sequences of rolls. If the seed
can be set, this is accomplished. Being able to set both the
seed and the counter does help to convince skeptics that the
position is played based only on the position and score, and
not, as some want us to believe, on some knowledge of future
rolls.
> >> 2- Whether a 5 digit subset (i.e. JF's counter)
> >> should be considered "small enough" for a 9 digit main set
> >> (i.e. JF's seed)?
>
> >I admit that I don't understand this question at all. It might be I'm
> >stupid, but I hope that's not the case.
>
> Actually the "9" above should be "14". The question
> was whether a subsequence of 99999 should be considered
> "likely" to occur more than once within 99999999999999.
The likelyhood of this is indeed very small. Indeed so small
that I an willing to stick out my neck and say that there is
no subsequence of 99999 rolls which occurs more than once in
a sequence of 99999999999999. Again see my earlier posts and
also Gary Wongs excellent post on random numbers. There are,
however, some questions I have in connection with that post,
but this is not the right forum for them.
> The key to whether I can elaborate on these is to
> first know whether some sequences repeat more than
> what would be considered "normal/expected". I already
> gave some hints as to how/why this may be significant.
> If I figured this one out, I may become interested in
> developing yet another hard-to-beat computer backgammon
> program myself... :)
There is one interesting point here. Let's assume that there
really is some anomaly regarding the dice rolls generated by
Jellyfish. Is there, theoretically, any way a neural net can
learn these anomalies, especially a neural net which doesn't
directly analyze the rolls? And if so, how great impact will
they have on the evaluation function which the net learns? I
have seen the answers to these questions, which I believe to
be the questions Murat really wants answers to, but couldn't
find them.
> Well, if I could analyse millions of games and find
> ways to make things particularly hard on you (vs. my
> program), this is not quite "cheating" ouright (i.e.
> spontaneously done as the situation requires, like
> rolling what you need, etc.) because it's done ahead
> of time and the player still has a "fair?" chance to
> overcome such difficult positions if he's "really good".
The thing is that a neural net like Jellyfish, does all this
by itself. In some way, it has analyzed millions of games in
order to learn how to make the best move in any position. By
making the best move in every situation, it is making things
particularly hard on its opponent. Unless there is something
wrong with the random number generator, a thing which can be
proven, and this anomaly will have significant impact on the
evaluation function, I cannot see how anyone could call this
manipulating, cheating or whatever.
> Now, such things may be possible and still not apply
> to JF. I don't think I'll be able to demonstrate anything
> anytime soon, but I think I'll succeed in at least
> making you guys wonder whether I'm not onto something
> for real. The alternative being, my having to accept
> the incredible reality that I may have been indeed a
> world-class player all along, without knowing it... :)
So far you haven't made me wonder if you're on to something.
And as a result of playing through the games you provided, I
don't think you are a world-class player, though undoubtedly
above average. However, since my FIBS rating is modest, only
1830 in 5000 experience points, I'm well aware that I can be
wrong in my analysis of your games. It is also possible that
you didn't play at your best when recording these games.
Regards,
Claes Thornberg
Murat Kalinyaprak
was a problem with my newsserver and the articles I posted
during the past couple of days don't seem to be distributed.
I'm reposting a few that I had not erased from my disk yet.
Claes Thornberg wrote:
> Murat Kalinyaprak writes:
>> Claes Thornberg wrote:
>>>I don't know what a significant deviation
>>>from the expected is, but if that sequence occurred more than, say 10
>>>times, I would worry about the randomness of a roll generator.
>> Yes, and this would apply to shorter sequences of only
>> 2, 3, 4 rolls also. Is there such thing as "inherently"
>> hard to play sequences of X rolls regardless of board
>> positions? If so, then any such high rates of occurrences
>> may be evaluated from a certain perspective...
> The shorter a sequence is, the more it will occur. Therefore
> the deviation from the expected will be larger. This applies
> only to absolute figures. In relative terms it should be the
> same. At least that's what I think. Could some correct me if
> I'm wrong, here or elsewhere.
I'm not concerned with some "any sequences" occuring more
often than expected. I was linking this to the question of
whether certain sequences are harder to play given "any"
board setup. For example, manually setup a board however
you want, then try to play "63 33 61", reset the board to
how it was, then play "63 42 55", reset the board to what
it was and play another 3-roll sequence, and so on. Repeat
the process for other board settings. At the end, could it
be possible to conclude, for example, that "63 33 61" causes
more blots/hits, creates more alternative moves to pick from,
leads to a certain type of game, etc. than "63 42 55"...? If
it's possible at all to draw such conclusions about a certain
50 sequences of 3 rolls and if you observe that you see these
sequences more often than normal, wouldn't you get suspicious?
Notice that this wouldn't effect the distribution of individual
rolls and 63, 33 and 61 could each occur equally and no more
often than 42 or 55. I'm not sure if my raising this issue
wasn't clear from what I wrote or if you ignored it.
> Not only is this method cumbersome and time consuming, it is
> also error prone. We will miss opening doublets, since Jelly
> will reroll such rolls without us ever noticing. It would be
> much better if Fredrik, for instance, could produce a set of
> rolls, which were sufficiently large, for anyone to analyze.
Not just any set of rolls though and JF is already capable
of producing 999999999*99999 rolls *consistently*. There
wouldn't be all that much special effort needed on the part of
Fredrick (by now I understand he is the lone creator of JF?)
to provide this.
> But I'm not sure if it's worth the effort to do this just to
> gain another JF admirer/advocate/defender.
Taking the section of JF that produces the dice rolls and
compiling it into a separate small program would be trivial.
Then, the user can enter, for example,
Starting seed = 1
For # of seeds= 500
and that little program would write out 500*100000 dice rolls
to a disk file. After those rolls are analysed, starting seed
may be set to "501" and the process may be repeated for another
batch of 500*100000 dice rolls.
> can be set, this is accomplished. Being able to set both the
> seed and the counter does help to convince skeptics that the
> position is played based only on the position and score, and
> not, as some want us to believe, on some knowledge of future
> rolls.
Being able to roll dice manually in combination with being
able to setup a board position manually accomplishes this.
"Seed/counter" doesn't add anything more to it. In other
words, if you take out the automatic dice rolling function
out of JF completely and restrict players to manual dice,
you could still make the above argument just the same.
>> Well, if I could analyse millions of games and find
>> ways to make things particularly hard on you (vs. my
>> program), this is not quite "cheating" ouright (i.e.
>> spontaneously done as the situation requires, like
>> rolling what you need, etc.) because it's done ahead
>> of time and the player still has a "fair?" chance to
>> overcome such difficult positions if he's "really good".
> The thing is that a neural net like Jellyfish, does all this
> by itself.
Whether it's called "neural net" or not, you can analyse
X billion games any wich way you want. For example, you
may make it analyse games that lasted less than 30 moves
or more than 80 moves, try to see patterns in them and
understand what made them so short or long. Then, possibly
you could induce a game shorter or longer than average,
(even though not as certainly as whenever you want).
> In some way, it has analyzed millions of games in
> order to learn how to make the best move in any position. By
> making the best move in every situation, it is making things
> particularly hard on its opponent.
Is everybody convinced that JF makes "the best move" (if
there is such a thing) in every situation...? Personally,
I'm not convinced of this. In fact, I'm wondering if the
so called "best moves" may not be different for real-life
vs. JF's purposes (whether intentionally or not)...
> Unless there is something
> wrong with the random number generator, a thing which can be
> proven, and this anomaly will have significant impact on the
> evaluation function, I cannot see how anyone could call this
> manipulating, cheating or whatever.
Right. But (as your sentence starts with), "unless"...!
>> Now, such things may be possible and still not apply
>> to JF. I don't think I'll be able to demonstrate anything
>> anytime soon, but I think I'll succeed in at least
>> making you guys wonder whether I'm not onto something
>> for real. The alternative being, my having to accept
>> the incredible reality that I may have been indeed a
>> world-class player all along, without knowing it... :)
> So far you haven't made me wonder if you're on to something.
> And as a result of playing through the games you provided, I
> don't think you are a world-class player, though undoubtedly
> above average.
Well, maybe you are just not the "wondering type" of
person. What's the 3rd alternative you are offering,
then? That, after 2000 games the score will eventually
be what would be expected from any "above average" player
against JF...? As long as there is even one question
still remaining unanswered, how can you say you are not
"wondering", unless you provide at least a "prediction"
type of an answer (i.e. the answer will prove itself
correct given time)...?
> However, since my FIBS rating is modest, only
> 1830 in 5000 experience points, I'm well aware that I can be
> wrong in my analysis of your games. It is also possible that
> you didn't play at your best when recording these games.
It was pretty much concluded that since the settings for
those 100 games were not correct, they can't be really
indicative of anything. I'll post the results of another
set of 100 games played with the correct settings soon.
Assuming that I get similar results, the dilemma is:
How can you, at the same time, claim that JF is world-class
player and someone who beats it is not...? It can't go in
both directions. Anybody who would prefer to judge my
moves to be less than world-class, has to accept the fact
that neither can JF be a world-class player, through the
simplest process of deductive logic. From the beginning,
I myself indicated that I wasn't making the best moves
(not best according to statistics and neural networks,
neither according to me) and it looks like this is apparent
to you, as it would be to any "good" player who looked at
the games I played against JG. What explanation can anybody
give, other than "luck", for a lesser player winning a good
number of games against a better player (let alone beating
it overall)? I had the impression that there was some
patters to JF's dice rolls, that I could predict and preempt
them by making purposefully "wrong" moves. If this is not
true, then I would really like to hear what you or anybody
else in this group would like to offer as an explanation
for what seems to be happening at this thus far...? I'm
not anxious to call myself a world-class player but I sure
would like to know what gives.
MK
Murat Kalinyaprak
Claes Thornberg
> I'm not concerned with some "any sequences" occuring more
> often than expected. I was linking this to the question of
> whether certain sequences are harder to play given "any"
> board setup.
I understand your question now, but there still remains some things to
consider, the most important being how you define "hard". Is it harder
to play a roll in a position the more alternatives you have, or is it,
as I'm inclined to believe, harder when the alternatives you have, all
leave a couple of blots?
> For example, manually setup a board however you want,
> then try to play "63 33 61", reset the board to how it was, then
> play "63 42 55", reset the board to what it was and play another
> 3-roll sequence, and so on. Repeat the process for other board
> settings. At the end, could it be possible to conclude, for example,
> that "63 33 61" causes more blots/hits, creates more alternative
> moves to pick from, leads to a certain type of game, etc. than "63
> 42 55"...? If it's possible at all to draw such conclusions about a
> certain 50 sequences of 3 rolls and if you observe that you see
> these sequences more often than normal, wouldn't you get suspicious?
Indeed I would get suspicious, but as I have pointed out in some of my
earlier posts, noone has ever showed that there is anything wrong with
the random number generator of Jellyfish. If you really want people to
take your arguments seriously, you must show that the generator really
has such a flaw, not merely suggest that it might be possible.
Furthermore, you must show that such an anomaly theoretically could be
detected by the neural network when it learns to play backgammon. When
trying to do this, i.e. showing that such an anomaly could be detected
by the NN, keep in mind that the dice rolls are not, not explicitly at
least, part of the input to the NN.
> Notice that this wouldn't effect the distribution of individual
> rolls and 63, 33 and 61 could each occur equally and no more
> often than 42 or 55.
I would say that it would not necessarily effect the randomness of the
rolls generated, but for reasons pointed out by Gary Wong I much doubt
it. To make this really clear, I am sure that the anomaly you suggest,
would effect the distribution of shorter sequences as well.
> I'm not sure if my raising this issue
> wasn't clear from what I wrote or if you ignored it.
I'm sure it was your raising the issue, which wasn't clear enough, but
maybe you don't agree. I haven't intentionally ignored anything you've
written.
> Not just any set of rolls though and JF is already capable
> of producing 999999999*99999 rolls *consistently*. There
> wouldn't be all that much special effort needed on the part of
> Fredrick (by now I understand he is the lone creator of JF?)
> to provide this.
Why limit yourself to these numbers. The upper bound of the counter is
not 99999. The range of value you can enter is from 0 to 99999, but as
anyone can see for themselves, the counter will continue to 100000 and
above when you keep playing. So one question to answer is how long the
sequences should be. For all three roll sequences to occur the correct
number of times you must have a minimum of 139968 rolls. These must be
generated from the same seed, of course. To test the distribution with
some kind of confidence you probably need five times as many rolls.
> >> Well, if I could analyse millions of games and find
> >> ways to make things particularly hard on you (vs. my
> >> program), this is not quite "cheating" ouright (i.e.
> >> spontaneously done as the situation requires, like
> >> rolling what you need, etc.) because it's done ahead
> >> of time and the player still has a "fair?" chance to
> >> overcome such difficult positions if he's "really good".
>
> > The thing is that a neural net like Jellyfish, does all this
> > by itself.
>
> Whether it's called "neural net" or not, you can analyse
> X billion games any wich way you want. For example, you
> may make it analyse games that lasted less than 30 moves
> or more than 80 moves, try to see patterns in them and
> understand what made them so short or long. Then, possibly
> you could induce a game shorter or longer than average,
> (even though not as certainly as whenever you want).
And what is your point? Is it better to have shorter games, or longer?
Could you please explain?
You don't seem to understand how the learning procedure of a NN works.
There is no interference with the learning procedure of NNs other than
selecting what input to feed the net, and how to adjust weights. Other
than this, Jellyfish has only played itself. What ever it has learned,
is there for us to learn as well. And analyze is not a term I'd use in
connection with NN learning. The adjustment of weights is a mechanical
task, it has nothing to do with analysis. But, its learning comes from
all the games it played against itself.
> Is everybody convinced that JF makes "the best move" (if
> there is such a thing) in every situation...?
No, it makes what it thinks is the worst move in every situation. Then
it rolls exactly what it needs to get away with it. Sorry, I could not
resist. I'm sure that it selects the move with the highest equity, but
remember that this move isn't selected from a list of all moves. Jelly
evaluates all possible moves on level five, but only some of these are
evaluated on level six and higher. So it's possible that the best move
doesn't get through from level five. The time factor might effect this
too. But in principle it selects what it regards as the best move.
> Personally, I'm not
> convinced of this. In fact, I'm wondering if the so called "best
> moves" may not be different for real-life vs. JF's purposes (whether
> intentionally or not)...
In real life you play opponents which might not be perfect. Then it is
possible to make plays, which leads to situations which you can handle
better than your opponent, or in which your opponent is likely to make
wrong cube decisions, etc. However, Jellyfish plays as if you a player
which is as perfect, or imperfect, as itself.
> > Unless there is something
> > wrong with the random number generator, a thing which can be
> > proven, and this anomaly will have significant impact on the
> > evaluation function, I cannot see how anyone could call this
> > manipulating, cheating or whatever.
>
> Right. But (as your sentence starts with), "unless"...!
Right, and so far, neither you, nor anyone else has shown any proof of
something being wrong with the dice generator.
> > So far you haven't made me wonder if you're on to something.
> > And as a result of playing through the games you provided, I
> > don't think you are a world-class player, though undoubtedly
> > above average.
>
> Well, maybe you are just not the "wondering type" of
> person. What's the 3rd alternative you are offering,
> then?
I never offered a third alternative, as far as I recollect.
> That, after 2000 games the score will eventually
> be what would be expected from any "above average" player
> against JF...?
Yup, for once you got me right. That's exactly what I expect.
> As long as there is even one question still remaining
> unanswered, how can you say you are not "wondering", unless you
> provide at least a "prediction" type of an answer (i.e. the answer
> will prove itself correct given time)...?
Can you please state what question you think remained "unanswered". It
would make it much easier to discuss with you if you did this clearly,
and succintly, and not obfuscated by vague speculations which are hard
to argument against.
> It was pretty much concluded that since the settings for
> those 100 games were not correct, they can't be really
> indicative of anything. I'll post the results of another
> set of 100 games played with the correct settings soon.
As pointed out by others, you have to play more than 100 games against
Jellyfish in order to prove anything.
> Assuming that I get similar results, the dilemma is:
> How can you, at the same time, claim that JF is world-class
> player and someone who beats it is not...?
Because one hundred games is not enough to prove anything. Please read
these posts about how many games you have to play against Jellyfish to
prove, with a certain confidence, that you are better than Jellyfish.
> It can't go in
> both directions. Anybody who would prefer to judge my
> moves to be less than world-class, has to accept the fact
> that neither can JF be a world-class player, through the
> simplest process of deductive logic.
How do you reach this conclusion? Because you won more points than JF?
Because you won more matches? I know I'm repeating myself, but I think
it is important to point out that 100 games is not enough.
> I had the impression that there was some
> patters to JF's dice rolls, that I could predict and preempt
> them by making purposefully "wrong" moves. If this is not
> true, then I would really like to hear what you or anybody
> else in this group would like to offer as an explanation
> for what seems to be happening at this thus far...?
Can you explain what you mean by "seems to be happening"? You have not
shown any proof that you can predict the rolls. So, what do we need to
explain? It seems to me that you are more interested in speculating in
the defectiveness of Jellyfish random generator, rather than trying to
prove yourself right, or wrong. You do not want to understand the most
basic thing about statistics, so how can we possibly argue with you. I
think you are one of those people who stick to an idea no matter what,
just like those who keep claiming the FIBS dice are rigged, etc.
Please, please, please. Give us some evidence that one single thing of
what you say is true. Try forming hypotheses, and then devise tests to
either reject them or increase your belief in them.
> I'm
> not anxious to call myself a world-class player but I sure
> would like to know what gives.
What do you mean with this last remark?
> MK
Murat Kalinyaprak
> mu...@cyberport.net (Murat Kalinyaprak) writes:
>> I'm not concerned with some "any sequences" occuring more
>> often than expected. I was linking this to the question of
>> whether certain sequences are harder to play given "any"
>> board setup.
> I understand your question now, but there still remains some
> things to consider, the most important being how you define
> "hard". Is it harder to play a roll in a position the more
> alternatives you have, or is it, as I'm inclined to believe,
> harder when the alternatives you have, all leave a couple of
> blots?
I had purposefully avoided defining "hard" and instead
left it at just mentioning a few possible elements that
could be part of a definition, but I agree with the last
part of your above sentence.
Before I go on, let me say that I appreciate your trying
to listen/understand the issues I'm raising and making
an effort to reply to them. This makes it easier for me
to justify spending more of my time in trying to make
myself understood. Later in this article, I'll be giving
a lot of numbers/stats which you may find interesting.
At this stage, let's also address another issue here. In
discussing these subjects, I feel more and more between
the rock and the hard place. I guess I'm just a curious,
analytical type and I feel that I should be able to express
my impressions/opinions/questions freely on these subjects.
At the same time, I don't want to come accross as I were on
a mission to prove that JF cheats or whatever else. So let's
continue this discussion with this understanding (i.e. in
hypothetical arguments, seeking answers/explanations to what
we seem to observe, etc.)
>> settings. At the end, could it be possible to conclude, for
>> example, that "63 33 61" causes more blots/hits, creates
>> more alternative moves to pick from, leads to a certain type
>> of game, etc. than "63 42 55"...? If it's possible at all to
>> draw such conclusions about a certain 50 sequences of 3 rolls
>> and if you observe that you see these sequences more often
>> than normal, wouldn't you get suspicious?
> Indeed I would get suspicious, but as I have pointed out in
> some of my earlier posts, noone has ever showed that there is
> anything wrong with the random number generator of Jellyfish.
The question is whether anyone has looked for flaws but
nobody found any, or nobody found any because nobody
looked for flaws...?
> If you really want people to take your arguments seriously,
> you must show that the generator really has such a flaw, not
> merely suggest that it might be possible.
I previously expressed that I would be curious/willing
to look at it if it was convenient to do so (i.e. if a
stand-alone version of the dice generator was available
to produce dice rolls in large batches).
> Furthermore, you must show that such an anomaly theoretically
> could be detected by the neural network when it learns to
> play backgammon.
I don't know that much about neural networks and I'm
not necessarily linking this to it. Let's say that it
could be a result of an unintentional flaw or a product
of old-fashioned statistical analysis.
>> Not just any set of rolls though and JF is already capable
>> of producing 999999999*99999 rolls *consistently*. There
>> wouldn't be all that much special effort needed on the part of
>> Fredrick (by now I understand he is the lone creator of JF?)
>> to provide this.
> Why limit yourself to these numbers. The upper bound of the
> counter is not 99999. The range of value you can enter is
> from 0 to 99999, but as anyone can see for themselves, the
> counter will continue to 100000 and above when you keep playing.
This is one of the first things I tested and discovered
for myself (and mentioned it in a previous article). It
is a reason enough to question why the user is limited
to entering 5 digits while the program goes beyond that,
but otherwise this is not critically important. All one
needs is let's say X million dice rolls generated by
entering one single seed number or a range/batch of seed
numbers.
> So one question to answer is how long the sequences should be.
> For all three roll sequences to occur the correct number of
> times you must have a minimum of 139968 rolls. These must be
> generated from the same seed, of course. To test the
> distribution with some kind of confidence you probably need
> five times as many rolls.
I have no problem with you or anybody else suggesting
how long the sequences should be or what the minimum
total numbers of dice rolls should be examined.
>> Whether it's called "neural net" or not, you can analyse
>> X billion games any wich way you want. For example, you
>> may make it analyse games that lasted less than 30 moves
>> or more than 80 moves, try to see patterns in them and
>> understand what made them so short or long. Then, possibly
>> you could induce a game shorter or longer than average,
>> (even though not as certainly as whenever you want).
> And what is your point? Is it better to have shorter games,
> or longer? Could you please explain?
I will try further down with some numbers/stats. But
don't necessarily expect a definite point or proof.
I'm just making some observations and asking what may
be possible, etc. As I joked about it before, as a
result of these discussions, I may end up developing
the best backgammon program yet (non-cheating of course:).
One thing you should notice here is that over the past
couple of weeks, I constantly adjusted down my estimates
on the length of an average game, all the way to 30. I
think the reason will become clearer a little further
down.
> You don't seem to understand how the learning procedure
> of a NN works.
I'm not linking it to NN necessarily. Note that I had
said "Whether it's called neural net or not" and I was
talking about simply analyzing X billion games, etc.
>> Is everybody convinced that JF makes "the best move" (if
>> there is such a thing) in every situation...?
> No, it makes what it thinks is the worst move in every
> situation. Then it rolls exactly what it needs to get
> away with it. Sorry, I could not resist.
No problem. Putting "the best move" in quotes and the
comment in paranthese next to it was supposed to spare
me from this, but I guess it didn't... :(
> I'm sure that it selects the move with the highest equity,
> but remember that this move isn't selected from a list of
> all moves. Jelly evaluates all possible moves on level
> five, but only some of these are evaluated on level six
> and higher. So it's possible that the best move doesn't
> get through from level five. The time factor might effect
> this too. But in principle it selects what it regards as
> the best move.
This is one area where we still don't seem to see each
other's point but maybe in time we will.
>> Personally, I'm not convinced of this. In fact, I'm
>> wondering if the so called "best moves" may not be
>> different for real-life vs. JF's purposes (whether
>> intentionally or not)...
> In real life you play opponents which might not be perfect.
> Then it is possible to make plays, which leads to situations
> which you can handle better than your opponent, or in which
> your opponent is likely to make wrong cube decisions, etc.
> However, Jellyfish plays as if you a player which is as
> perfect, or imperfect, as itself.
Personally, I believe computer programs like JF are
more vulnerable than humans. Human opponents could
possibly *mutually* adjust (or at least try to adjust)
to each other's styles after a small number of games.
JF can't do that. After playing a few dozen games
against it, for whatever "unfounded":) reasons, I may
make certain types of moves and if those moves work
well enough consistently enough, I can consider them
as "best moves" as far as playing against JF. In other
words, my "best moves" may be flexible and vary from
opponent to opponent. But JF's best moves don't vary.
That's why, in a situation like this, JF supporters
have no other option than arguing that it will require
2 billion 1-point games to determine whether somebody
is better than JF. Do you see that a human may not
need to base his moves on statistics spanning millions
of games, but can "neural network train" himself in as
little as a few dozen or few hundred games? That's why
I keep stressing that it's possible that playing the
JF way will not work against every human, and JF may
very well be beaten more easily by playing non-JF way
against it (but the moves that work against JF may
fail big time against another human).
>> Right. But (as your sentence starts with), "unless"...!
> Right, and so far, neither you, nor anyone else has shown
> any proof of something being wrong with the dice generator.
Well, I just did a few days ago. Articles posted to
this group don't seem to get disseminated very well.
If you haven't seen the ones I posted about JF's dice
bugs, maybe you'll see them later. At this point, I'm
not sure about the whole extent of implications of the
bug/s discovered. I asked Fredrik to further clarify
certain things and we'll see if/how he responds.
>> That, after 2000 games the score will eventually
>> be what would be expected from any "above average" player
>> against JF...?
> Yup, for once you got me right. That's exactly what I expect.
I doubt that it will, but if it still means anything
to me to win another 1000+1 games against JF, maybe
I'll get to play that many games once the long/harsh
Montana-winter-nights arrive in a few months.
>> As long as there is even one question still remaining
>> unanswered, how can you say you are not "wondering", unless
>> you provide at least a "prediction" type of an answer (i.e.
>> the answer will prove itself correct given time)...?
> Can you please state what question you think remained
> "unanswered". It would make it much easier to discuss with
> you if you did this clearly, and succintly, and not
> obfuscated by vague speculations which are hard to argument
> against.
Based on the results I got thus far, it still simply
boils down to either me being as good as JF (i.e. a
world-class player) or JF being as bad as me (i.e.
any above-average [per your terms] player). The only
other option left for anyone to cling to is that the
results will change over the long haul and this "one
way or the other" will no longer apply (which you
expressed above to be your expectation).
>> I had the impression that there was some
>> patters to JF's dice rolls, that I could predict and preempt
>> them by making purposefully "wrong" moves. If this is not
>> true, then I would really like to hear what you or anybody
>> else in this group would like to offer as an explanation
>> for what seems to be happening at this thus far...?
> Can you explain what you mean by "seems to be happening"?
That I seem to be beating JF at 15 point matches,
single games with gammons, 1-point games without
gammons, with manual dice, with automatic dice,
at the highest level settings... Since you and
some others argue that a few hundred games are not
enough to conclude anything, I have no choice left
but to use uncertain expressions like this.
> You have not shown any proof that you can predict the rolls.
I posted the results of a dry-run I had made and
mentioned many times that I was making pre-emptive
(or so I may be superstituously believing) moves
against JF that I thaught were helping me beat it.
What else proof could I offer for such a thing?
Would I bet my money on it? No. Many times I myself
voultarily added that it may all be my imagination.
It's up to you to make whatever you want out of it.
> So, what do we need to explain? It seems to me that you
> are more interested in speculating in the defectiveness
> of Jellyfish random generator, rather than trying to
> prove yourself right, or wrong.
I don't have to be right. But where do you start,
if not by following your own impressions? Sometimes
you just think you observe something but you can't
explain it. And in such cases, sometimes you never
get to find out for sure. For example, I had raised
an issue about whether JF's seed/counter numbers
were beyond their real scopes. I didn't really know
what exactly lead me to think that. Recently, Fredrik
indicated that JF picks seeds (based on system clock?)
smaller than 65535, because seed greater than that
produce invalid dice numbers (if I understood his
brief explanation correctly). If this ends up
validating my suspicion/speculation, it may be that
my impression was based on (subconscientuously?)
observing that JF was always picking small seed
numbers but that I was never able to put my finger
on it...
> You do not want to understand the most basic thing about
> statistics, so how can we possibly argue with you.
It's true that I'm not an expent on statistics, but
in just a few more lines I'll try to give you some
stats to the degree that my knowledge/capacity permits.
> I think you are one of those people who stick to an idea
> no matter what, just like those who keep claiming the
> FIBS dice are rigged, etc.
I disagree. I'm just a little slow to take people's
word for things, especially if they have not offered
themselves what they would expect from me.
> Please, please, please. Give us some evidence that one
> single thing of what you say is true. Try forming hypotheses,
> and then devise tests to either reject them or increase your
> belief in them.
Let's first make sure to stress the fact that I was
asking a lot of questions here and soliciting opinions
from others, not necessarily making definite claims
and/or trying to prove they are true. Now I'll give
you a bunch of numbers and would like to know what
you and other readers make of them.
I'll start by trying to set the foundation based on
known facts. Please feel free to correct me if you
find errors in the following.
The average value of a dice roll is 8.17 (i.e. if
we roll dice 36 times, get each possible conbination
once, add up the numbers and divide it by 36, that's
what we would get).
The minimum number of pips required to bear of all
your pieces without any contact with the opponent,
and with *zero waste* during bear-off, is 167.
The minimum number of rolls required to accumulate
167 pips is 21 rolls (i.e. 167 divided by 8.16 =
20.47 rounded up).
Assuming the above are correct, let's look at the
last 100 games I played against JF, with these in
mind:
# of games ending in
rolls that many rolls
----- ---------------
18 3
19 1
20 5
21 15
22 6
23 10
24 10
25 6
26 7
27 4
28 7
29 6
30 2
31 2
32 1
33 1
34 1
35 2
36 2
39 1
40 1
42 1
43 1
45 1
47 2
49 1
53 1
Now, remembering the significance of 21 rolls, if
we count the games that took 21 rolls or less we
come up with 24 games (almost exactly a quarter of
the 100 games). Do you think this can be evaluated
to be significant in any way...?
With only a few pips wasted during bear-off, a
non-contact game can very easily last another
1-3 rolls. Notice that 50% (half!) of the games
ended in 24 rolls or less... Do you think this
may be indicative of anything...? Do these numbers
look like they could reflect real-life statistics
to you...?
In all but 3 of those 24 games lasting 21 rolls or
less the winner was hit from 1 to 3 times, but the
average roll value never dropped below 8.05, and
reached as high as 10.86... Overall average roll
value for all those 24 games was 9.40. Doesn't these
look like quite generous dice rolls to you...?
Should we also consider that the rolls need to be
happening in a certain fashion for one side to go
on "auto-pilot" (i.e. build a 5-block in 3 rolls,
hit a couple of pieces and bear-off the last few
pieces while the opponent only gets to watch the
"show")...? Rolling the wrong big numbers at the
wrong time or too many times in a row can do more
harm than good. So, besides the dice being a little
generous, may be it also needs to follow sequences
suitable/leading to short/fast games...?
Please don't get the impression that I'm crying
about JF beating me with such quickies. In fact,
it's the other way around. I won 16 out of those
24 games. In game #35 which I also won, the total
dice value I had rolled was only 169, merely 2
more than the minimum needed for a non-contact
game with a perfect bear-off. I guess I wasn't too
bad at squeezing the juice out of each number...
The average number of roll for all 100 games was
merely 26.68! Maybe I should revise again my last
estimate of 30 rolls as average length of a game.
If this doesn't reflect real-life, does anybody
have an idea about how many rolls a game lasts on
the average in real-life...?
Do these numbers have any meaning/value for you?
Please feel free to give me whatever feedback...
>> I'm not anxious to call myself a world-class player but
>> I sure would like to know what gives.
> What do you mean with this last remark?
I mean that if all the expectations point in a
certain direction but the results/facts don't,
then there must be some kind of explanation for
it. (I know, I know, 100 or 200 game results are
not enough:)...
MK