Those Darn Dice

[emdoub]

Okay - everyone has heard complaints, often from themselves, about the FIBS dice.  What we're told is that if we suspect the dice here, try playing gnubg - if you think gnubg's bot is manipulating the dice, set it so you enter the rolls and see that it's a strong player anyway.

Me, I figured that the dice here were whimsical - they're either on your side, or against you, at random - so it's fair between players.  But I got tired of the high number of "interesting" rolls, so I went off for a while to play against the gnubg bot offline.  

Gnubg cheats.  I got annoyed enough to get a pad of paper and keep track of rolls - specifically, the number of times we were each unable to get off the bar with 5,4,3,2 or 1 points open versus number of times sent to the bar, number of times able to hit a single blot on the home board while coming off the bar, and number of large doubles (4-4 or greater) after either side started bearing off or all markers were behind the opponent.  

At 50 games, the numbers were showing a pattern.  After 100, the pattern became significant.  

Yeah - the gnubg bot is a strong player - I won't argue that.  But what I've never seen is someone from the coding team at the gnubg project, or someone in admin here, say that the dice were entirely random, unweighted by anything.  Easy to say, but I've never seen it said.  What is said is misdirection - no outright "the dice are random" statements - but why will nobody say that?

I do fairly well with RL dice.  I do pretty good here, now and again.  What I want is random dice - not weighted in my favor, not weighted against me - no matter how evenly that weighting happens.  Is that available anywhere?  Is that even claimed to be available anywhere?

[Chris]

Could you zip the match-files and post them here?

[emdoub]

Could you zip the match-files and post them here?

I'm not even sure of where the match files are stored, and, in playing against a 'bot, I occasionally succumb to the temptation to nuke the program when it gets egregious - it's not like I'm hurting anyone's feelings or rating like that.

But that wasn't the point I was trying to make.  My statistical methods are, admittedly, pretty crude - hatchmarks on paper, for particular situations.  I've no wish to _prove_ to anyone that the GnuBG bot cheats at dice - I'm happy to be taken as a whining ranter on that.

What I'm looking for is someone who's got access to the code to tell me that I'm wrong - that the dice here or with GnuBG are actually, honestly random.  I've never seen that claimed, and I'm wondering why.

[Chris]

The code is open source - everyone can read it.

Here is the source file where the dice are rolled:
http://cvs.savannah.gnu.org/viewcvs/gnubg/...e.c?view=markup

GnuBG does not cheat.

[emdoub]

Thanks - that's what I was looking for.  I'm not a coder, and while 3dFibS creates a directory for the GnuBG code, it leaves it empty - so now I can try to look and see for myself.

If I'm simply being outplayed, I'll deal.  If my luck is just bad with the dice these days, I'll deal.  Being told straight-up that the dice are honest does help - but when I need a 5 or 6 to get off the bar, and GnuBG needs a 6 to get out from behind my block, and 4 turns later, I still can't roll higher than a 4 and GnuBG has rolled boxcars twice, it's easy to get suspicious - and disgusted.

[emdoub]

Okay - I haven't done any coding since I was using dBase - and C has never been one of my skills - so I could be misreading this code.  It *looks* like there's several different possible sources for dice, including a dice file, random.org, MD5, and some other stuff that's more obscure than I can figure out.  

If I knew what I were doing, I'd have a better understanding of why there are so many different options for random numbers - and I might even find where the routines are called both when GnuBG and player roll dice.  Or even where the code is identified as being the random-dice routine library used by all GnuBG implementations.

However, I do have Chris, who seems to know the code/program used by GnuBG, who tells us straight-out that GnuBG uses the same random number generator that the person it's playing does - so I'm closer to what I wanted than I was when I started this thread.  Thanks, Chris.  

[gammboy]

Okay - I haven't done any coding since I was using dBase - and C has never been one of my skills - so I could be misreading this code. It *looks* like there's several different possible sources for dice, including a dice file, random.org, MD5, and some other stuff that's more obscure than I can figure out.

What you see there are the various choices for random number generation, and in the case of the file, an option to read the dice from a file you supply.  Random number generators are not created equally.  Gnubg gives you the option of various methods of dice generation.

You can believe that, since the source code for gnubg is free and subject to public scrutiny, any cheating by the program would be loudly and widely protested.  It could not possibly be kept secret.  

Besides that, since the program is open source and free,  there is no motivation for the coders to have the program cheat.  Their purpose in creating and improving gnubg is to make a program that is very good at playing backgammon.  

Though I am not nearly the coder that Chris is, I will concur with his assertion that gnubg does not cheat.

On the subject of fibs, the source code is NOT available.  And so, there is no way to definitively say that it does not choose favorites in the way that you describe.  However, we are told that there is a SINGLE random number generator, and it  hands out numbers a round-robin fashion amongst all in-progress games.

This means that, even if you knew the method for random number generation, and you had the seed for that random number engine, you would still not be able to predict the sequence unless yours was the only game in progress, which is almost an impossibility.

And, needless to say, since the dice are generated by the server, there is no opportunity for dice manipulation on the client side.  i.e. Neither you, nor your opponent can influence the dice.

You mention that you use 3DFibs.  It stores match files for every match you play.  Those can be directly imported into gnubg for analysis.  They are also simple text files that you can read to do your own dice analysis of the games you have played, if you wish.

Finally, the fibs server has a command that will display for you its own analysis of the dice it has generated.  The actual command escapes me right now, but if you type 'help' at the command line, you can find it.

The issue of dice comes up from time to time.  The answer generally is: gnubg does not cheat, and the fibs server probably doesn't cheat.  In the case of gnubg, you (and everybody else) has/have the code.  In the case of the fibs server, only those who have seen the code know for sure, but where's the motivation to cheat? (psychological experiment theories aside)  

GB

[Chris]

Or even where the code is identified as being the random-dice routine library used by all GnuBG implementations.

As gammboy said, there are several implementaions. The default one, the Mersenne Twister, is found in:
http://cvs.savannah.gnu.org/viewcvs/gnubg/...r.c?view=markup

You also can change the settings to manual - so you can roll the dice yourself and reroll if you don't like them.  ^_^

The question about cheating bots is a rather regular one, see:
http://www.bkgm.com/rgb/rgb.cgi?menu+computerdice with articles like
http://www.bkgm.com/rgb/rgb.cgi?view+1152

Regards, Chris

[claudio_first]

Hi,
Emdoub you are wrigth only when you say that the dices here are.. "......." put inside what you want

Gnu doesn't cheat, I have Him, set up supreme and world class and he's my sparring partner for 7/11/15 pts matches; sometimes (sometimes.. very sometimes) I win 'cause he, with the neural net, is so clever that even a 2-1 is good!

About the random generation we can talk forever.. In our Team (euit) we have a lot of discussions about and have a look also HERE, made by the German Team Captain, is all about Zone's dices..

May be is better say that here we have to play in a different way than on Playmaker or MSN Zone or Games Grid.

Here really are too much doubles (ex. 3,4,5 even 6 at the bearoff or 6,7 in a game for both), the same dices (ex. 3-2) three/four times consecutively, if you have low experience you are mure luck than an old player (try with a new nick;), here I say yes and here I say no ) and the way of playing is less assaulting than in another server.

Cheat the dices.. is possible :angry:
About 2 years ago on Zone, a player (with +members, backgammon team menagers and a lot of other players)  did that to demonstrate, but always, as the player told, a little window will open (on the up wrigth corner, if I remember well) so you will know!
After that I don't know if somebody else did that but.. Windows is Windows open as its name

Claudio

[emdoub]

The question about cheating bots is a rather regular one, see:
http://www.bkgm.com/rgb/rgb.cgi?menu+computerdice with articles like
http://www.bkgm.com/rgb/rgb.cgi?view+1152

Regards, Chris

Good article, there - thanks for the pointer.

Okay - my statistical survey didn't show cheating with the dice - it just shows me getting my butt whupped, but the play is clean.  I can deal with that.

It's probably still going to irk me that I get half of my 6-6 rolls when I'm sitting on the bar, but having someone who knows state that the dice aren't really rigged will keep my paranoia down to reasonable limits.  It's not that the dice are unfair; it's that the ghods like tormenting me.  

I've played a *lot* of money games, over the decades - I guess that doesn't make me world-class, I've got some learning to do.  Oh, well - my ego is accustomed to being stepped on, I'll live with it.

Thanks, all, for the helpful comments and pointers - it greatly reduced my frustration level here.  I'll see y'all around the playing tables, I hope.

Midnight Writer
which shortens easily to MW
which shortens even further to emdoub, if you say them both aloud.

[claudio_first]

Another thing..
growing up your/our experience we look with more attention to the dices we get.. as a beginner everything can be good but now no! We all are thinking before a move and about what our opponent gets? We are more critical..

See you at one table with pleasure

Claudio

[gammboy]

Cheat the dices.. is possible
About 2 years ago on Zone, a player (with +members, backgammon team menagers and a lot of other players) did that to demonstrate...

On zone.com, cheating the dice was possible because of what was, in my opinion, a design flaw of the software.  They designed their system such that the CLIENT generated the dice and sent the roll back to the server.  This allowed the clever hackers to manipulate the roll and, in essence, pick their own dice.

On fibs, this technique will not work, because the SERVER generates the dice, and sends them to the client.  The server then enforces legal moves based on the dice IT picked. There is no opportunity for the client to change the roll.  Therefore, dice manipulation on fibs is not possible.

GB

[claudio_first]

..... On fibs, this technique will not work, because the SERVER generates the dice, and sends them to the client.  The server then enforces legal moves based on the dice IT picked. There is no opportunity for the client to change the roll.  Therefore, dice manipulation on fibs is not possible.

GB

Sure about that, was on Zone.

Claudio

[Hardy_whv]

It's probably still going to irk me that I get half of my 6-6 rolls when I'm sitting on the bar ... 

That's what called selective memory. I bet a 250ltr cask of the finest scottish whiskey, that you get much more double six when not on the bar!

I just analyzed my last 35 11-pointers against GNUBG. GNUBG had 140 66, I had 141. Of GNUBG's 66, 32 where from the bar, dancing (=22.8%). Of my 66s 28 where from the bar, dancing (=19,8%). I guess your 66-from-the-bar-ratio is in that area as well.

I'd recommend that you save all your matches against GNUBG and also analyze them. When you play with the same error rate as GNUBG you will win 50% of all matches. (For my training matches against GNUBG I put the settings of the evaluation to 0.1 noise. So GNUBG plays less than perfect, average error rate nearly -1.5. I sometimes wish I'd put the noise to a higher level even.)

Hardy    

[dorbel]

What everybody wants is random dice and that is what everybody gets, on fibs, on gnu and so far as I am aware, on every reputable server. However, when they ask for random dice, some people mean something subtly different; what they want is the luck spaced out fairly evenly. This just isn't possible and arguably isn't even desirable! Luck, in a raw form, can be studied by the bots, listing all the possible rolls on each turn and then estimating the equity resulting from each if the roll is played in the best possible way. If you begin to look at luck rates carefully, you will see that the vast majority of games feature a luck rate for one side that cannot be overturned by superior play. In money play, more than 90% of the games are decided by the dice and the major skill is actually making sure that the game finishes with the cube at the right level to maximise your return from the game. This incidentally, is why some players survive quite well in the weekly chouette even though they make obvious technical checker errors. Occasionally there is a game that will be decided by skill, but they are much rarer than people imagine. Big luck swings are the norm, live with it and love it!

[Hardy_whv]

If you begin to look at luck rates carefully, you will see that the vast majority of games feature a luck rate for one side that cannot be overturned by superior play. In money play, more than 90% of the games are decided by the dice and the major skill is actually making sure that the game finishes with the cube at the right level to maximise your return from the game.

Hi Paul,


I fully agree with your statement on the randomness of dice.  

One comment on the importance of luck vs. skill: I do not play money game very often, so I can't say much about the relevance of luck and skill in a single game. But if it comes to match play, I think, things change dramatically. I have a nice collection of 230 top backgammon matches. Of most of the matches at least one of the players is a Giant of Backgammon. The average match length of those matches is 17.2. I analyzed the Snowie error rate for all matches. When one players had an error rate 2.0 better than his opponent, only 21.2% of the matches were won by the weaker player. If the error rate differs at least 4.0 only 6.1 matches were won by the weaker player.

Lets say a world class backgammon player plays at an error rate of approx. 3.0, an advanced player with an error rate of 7.0. Then 93.9% of all matches (with an average match length of 17) will be won by the world class player.

I think that's enough motivation to become a good player and to believe in the importance of the skill factor  

Hardy  

[dorbel]

I won't argue with your statistics or the conclusions that you draw Hardy. However there will be many backgammon players, the enormous majority in fact, who don't play a 17 pointer from one year's end to the next. All British tournaments, almost all online tournaments are composed entirely of much shorter matches, where there just isn't time for the luck to even out. However, there is obviously something to be gained from improving your play, even if it is only for the satisfaction and interest! As far as single games go, try playing 100 games against a bot. Even though the bot will outplay you 90 times out of 100 (or better), it will do well to win 55 of the games and often won't get to that.
One factor that nonbot players like to trumpet is that there is more to the game than playing pure bot plays and they do have something. You have to know what the strongest plays are of course, because they will represent your best chance against the best players, but when playing somebody considerably weaker, selecting slightly weaker but more complex plays can pay dividends. This is even more true with the cube. To give just one of many examples, I know a well known British player who never gives a four cube until it is a cash. This means that I can double him in a little earlier than usual, because I know that he will waste most of the value that he should be getting from cube ownership. These areas, complicating or simplifying the game to order offer at least as much scope for improving that elusive win rate as does reducing our error rate.

[emdoub]

Upon careful review, it has become apparent - GnuBG's 3dFiBs implementation is simply remarkably lucky.  It can put me on the bar with 2 points covered and a divot on its home court, and watch me spend 4 rolls or more on that bar.  It can, reliably, come off of the bar the first roll when I have 4 or 5 of my home points covered.  I've seen it play to move points en masse rather than run them in to be able to cover a vulnerable point outside of home, or on home, while I'm unable to come off of bar.

Has playing against it improved my game?  Yeah, I suppose it has.  Am I looking for an implementation in which I can roll the dice because I don't trust the program to do so honestly?  Yup - I am.  Perhaps an implementation which would give me the option of analyzing my games would change my mind - but I don't have one, and, frankly, I don't much give a rip.  I've seen too many strings of rolls that stretch the imagination - very few of them in my favor.  I have, indeed, been simply outplayed when the dice seemed random - but I've also seen an awful lot of very lucky dice, and seen play that seemed to anticipate those dice.  If it were human, I'd want it rolling with my dice, not its own.

Thanks for the input, folks - I believe that none of you believe that GnuBG or FiBs is rigged.  I, personally, believe that they are - with the dice favoring people randomly on FiBs, but GnuBG playing as if the dice were on its side - usually, to good result.

Hardy_whv - I've a passing fondness for Glenfiddich or Dalwhinnie (though I prefer the Tyrconnell or Black Bush to either) - and that's a dangerous bet.  I logged, with my usual pencil tally marks, every boxcars I've rolled in the last week.  24 marks in the "oh, great" or "not ideal, but I can use 'em" category, and 37 in the "what an awful time for boxcars" column.  In the "now, it's a race, nobody has to worry about blots" tally, I've come from behind to win with good rolls 8 times - GnuBG has done so 14 times.

Mayhaps I'm just a bitter, twisted old man (well, I probably am), but I've played thousands of RL games without any worries about the dice being rigged - and can't get over how consistently lucky GnuBG is with the dice.  I don't mean 'can do something worthwhile' with a mediocre roll - I mean gets the only roll that'll do it any good far more frequently than I've seen when the dice make noise on a table.  

It'd probably be good for me to learn C well enough to look over the code for GnuBG - for reasons other than settling my paranoia playing backgammon.  But I probably won't - I'll just look for an implementation in which I can roll the dice for both sides, and *know* they're honest.  I'll still get outplayed - but somehow, I suspect that I won't be so consistently out-lucked.

[socksey]

Much has been discussed on the randomness of Fibs dice.  Here's another thread on this subject from the past that you will no doubt be interested to read:

http://www.fibsboard.com/index.php?showtopic=742

And, of course, this link should be required reading for all newbies to Fibs:

http://www.alef.co.uk/fibs/archive/dysfunction.html

Enjoy!

socksey



"I wish men had boobs because I like the feel of them. It's so funny - when I record I sing with a hand over each of them, maybe it's a comfort thing." - Emma Bunton, Baby Spice of the pop band Spice Girls

[emdoub]

Final evidence, I guess....

Playing against GnuBG tonight, I saw some play that dropped my jaw.  I had two blots on the GB home court, and 3 empty spaces on my own home board - everything else filled there.

GnuBG rolled, and hit both of my blots - leaving divots on it's home board.  3 points covered, one empty, and two divots.  Remarkably bad play, from everything I've ever learned about the game.  It's outer board was left a mass of divots - 4 divots, one point covered, one empty.  A suicidal move - getting one off the bar onto either of those divots would have been remarkably dangerous for it, getting both of them off the bar would have been more so - even if I only hit one divot.

Of course, I couldn't make it off the bar, and on its next roll, GnuBg was able to cover both of those vulnerable points from the divots it had in outer board.  Remarkably successful strategy, but too risky for standard (or match-grade) play - unless you know you have the dice in your pocket.

I'm tired of "interesting" dice.  I've taken to offering the cube when a gammon is almost certain, just because the dice on FiBs are being one-sided in my favor.  I've seen great runs of dice happen in RL - but they're nowhere near as frequent there as they are on FiBs, or playing against GnuBg.  

I can understand some folks, trying to come up with a backgammon program, who want it to be hard for meatpeople to beat, giving it something of an edge with the dice.  I can even understand someone, creating a playground for backgammon lovers, nudging the dice so that they're more 'interesting' than actual, random dice.  What I can't do is approve of either.

Anyone know of a BG-play site that's got honest dice?  I don't like when FiBs cheats - in my favor, or against me.  I just want honest dice.  I don't need good dice every time, and I'm sick unto death of amazing dice - against me or for me.