I use gnubg, and there have been some moments where I wanted to pull my hair out at some of the ridiculous rolls that occur in my games against it. No, this is not a "my bot cheats" thread, because there have been plenty of times when I've gotten quite lucky against gnu as well. But I do feel like too many doubles are thrown with the computerized dice, and a lot of games are settled on which player gets the luckiest throws. The main scenario that seems to come up a lot is that when the game comes down to a straight race at the bearoff, one player will throw multiple doubles to win the match.

I suspect that this is a case of selective memory, where these games have had the biggest impact on me emotionally, and so I tend to remember them more. But I also know that in my live games, it seems that the winner is, more often than not, the player who out-played his opponent, and not the player who got the ultra joker to win the match. So I started playing gnu with manual dice just for fun to see what would happen. It turned out that I was achieving much better error rates and winning more games. So I decided that I should start documenting my experience with this so I can keep some long-term results.

So to start this experiment, I decided to play 10 games in a row against gnu. I would play 3 with manual dice, 3 with auto dice, then alternate the last four games. The results blew my mind! Here are the results for the first 10 matches, with the dice type used, the winner of the match, and my snowie error rate:

Game 1: manual, gnu, 4.9
Game 2: manual, gnu, 5.0
Game 3: manual, me, 6.1
Game 4: auto, gnu, 46.7
Game 5: auto, gnu, 38.7
Game 6: auto, gnu, 19.0
Game 7: manual, gnu, 6.0
Game 8: auto, gnu, 36.8
Game 9: manual, me, 14.8
Game 10: auto, gnu, 31.0

I'm stunned, and I honestly can't explain it. My first thought was that maybe when I use gnu's dice I play more quickly than usual and make dumb mistakes, whereas taking the time to roll the manual dice each time slows down my game and causes me to analyze a bit more before I make my decision. I will also say that the 14.8 on game 9 was due to a very bad play I made, while I was flawless on every other checker play in that game. If I had made the correct move there that error rate would have been closer to the others.

I will also say that I noticed more doubles being thrown in the auto dice games, although there was one manual dice game that had quite a few doubles as well. Is there anything documented that shows that dice algorithms tend to come up with more doubles than they should?

I still don't think the bots cheat or the dice are fixed or any of that foolishness, but it's clear that I play better when manually rolling. Any thoughts? Is this all just psychological? Did I get those awful error rates on the auto dice games because I EXPECTED to play poorly with the auto dice?

How do you calculate the error rate anyway? Is there a range of error rates that are considered good or bad?

In gnubg, after I analyze the game, I choose "Match or session statistics" from the analyze menu, and it gives you all your stats, including your snowie error rate. As for what the snowie error rate says about your game, this link shows a nice little breakdown: http://www.bkgm.com/rgb/rgb.cgi?view+1333

This seems reasonable to me. I know I get clicky sometimes when playing and make mistakes that I wouldn't have made if I had took a second to look at the position. I also know I miss cubes when I'm clicking quickly.

Other information that would be useful to have:
* Checker/Cube error rates separately
* # of moves (it's easier to play with a low ER in a short game)

Although a pattern is clearly emerging, it would be interesting to see whether it continues over the next 10 games or so. 10 games of backgammon is a very small sample, and it's not completely implausible that this might just be a weird string of coincidences.

I'll throw in an heretical opinion here, just for the hell of it. If you google 'J B Rhine' you'll find the results of 30 years' experiments with dice. Some Russian universities have also done enormous studies. Why is there such a thing as 'beginner's luck'? Why is someone feeling positive and energetic more likely to roll 66? Some think there is a quantum effect at work, with the deeply non-random energy of a mind affecting, very very slightly, the random chaos of a dice throw...

I thought about including this, but wasn't sure that anyone would be interested in the first place, haha. I'll start taking that info down as well. I can tell you this much: All 5 of my manual dice games were longer games (in the 25-30 move range), and my auto dice games were shorter. My cube decisions were very strong in my manual dice games, and were weaker in the auto dice (along with my weaker checker play in those games). So I definitely play the cube better in the manual games as well. I'll start tracking these stats as well...maybe I'll put it all in a spreadsheet.

You've found the answer already: playing slowly and carefully (which the manual dice forced you to do) has a HUGE effect on your error rate.

I performed a similar experiment on myself some years ago. I played a set of matches against Snowie where I only played when I was alert, well-rested, and eager to play. I'd play only one match at a time, after which I would do something else.

I then played a series of matches under non-optimal conditions: I was tired, I didn't feel like playing, I had a cold, or I forced myself to play at blitz speed. My error rate in the second group was more than twice that of the first group. Now, when I go to a tournament, I'm even more careful to make sure I arrive at least a day early, get a good night's sleep every night, eat on a regular schedule, and focus carefully on each match. It makes a big difference.

HOT DIGGITY DAMN!!!

I've done the same experiment, myself, but did not keep written record. I KNOW I win more (many more) games against Gnu with manual dice. I don't even think I play as well with manual dice because messin' with the dice, looking for the dice numbers on the screen and all that is distracting.

I was tempted to post about it, but decided not to. Y'all would assume it's just another amatuer's "Gnu cheats" thread, but NoChinDeluxe posted the evidence.

As far as I'm concerned, the jury is back. Verdict:

GNU CHEATS!!!!!!!!!!!!!!!!!!!!!

It seems to me that it's "cheat mode" is sometimes on; sometimes off. More on than off. The next time y'all are in a ridiculous losing streak against Gnu, try the manual dice for yourselves. You'll see:

GNU CHEATS!!!!!!!!!!!!!!!!!!!!!

Thanks, NoChin, I feel vindicated.

The use of manual dice is distracting, but the extra physical activity of sitting straight in my chair and shaking & rolling the dice may be stimulating to the brain, but that does not account for Gnu's "lucky" rolls over and over, nor my unlucky rolls.

I've noticed predictable patterns. For example, if the game is close and it's become a race to the finish line: Guess who gets doubles and who gets 12s.

GNU CHEATS!!!!!!!!!!!!!!!!!!!!!
GNU CHEATS!!!!!!!!!!!!!!!!!!!!!
GNU CHEATS!!!!!!!!!!!!!!!!!!!!!
GNU CHEATS!!!!!!!!!!!!!!!!!!!!!

An' don't nun o' yew nay sayers say nuthin' 'til you've tried it yerself!

AND FURTHERMORE,
Gnu may be a horrible teacher in some ways. When it hits almost everything you leave open, you learn to play too conservatively. (I think my posts in another recent thread illustrate that.)

When taking chances against Gnu, you are NOT taking chances...You are MAKING SACRIFICES.

You know, Bill, this is also a reason I don't enjoy playing online too much. I feel too compelled to play quickly, and with the added pressure of the quicker clocks that most online players like, it definitely has a negative impact on my game. The same thing happens to me with poker. I like the slower-paced environment of a live game, and I find it easier to focus for long periods of time.

I think I'll continue with my experiment. It will be helpful to me to keep stats records on all my practice games, and I think it will also help improve my focus. I'll post more about it in the future if I see any more interesting trends emerge.

This experiment says absolutely NOTHING about whether or not GNU Cheats.

The error rate comparison has nothing to do whether one player is getting good dice or bad dice. Even if the gnu was cheating the error rate would measure how well you played your rolls, good or bad and nothing else.

Your "predictable patterns" are certainly no more than selective recognition, the lucky rolls for GNU stand out because you notice them, and not all the average below average rolls that go into it. People that claim bots cheat are doing nothing more than giving themselves excuses for why they can't beat the bot. And of course, no one that claims that the bots cheat, can every produce a significantly size sample set and statistical analysis to prove it, they go on there "feelings," which when it comes to something like dice statistics are almost certainly wrong.

Here is a little experiment that one can do to see that GNU is not tinkering with the prng output.

Start gnu, go to the dice options and enter a seed number, make not of this seed number.

Play a session with GNU, without exiting the program. Print the match records for that session.

Now restart GNU and enter the same seed number, but when the first roll comes up select swap players and play the session that way.

You will see that no matter the position the exact same sequence of rolls is coming up! If GNU were cheating this would change.

No it does not. I know it's competitor of ours, but hopefully that will put more weight in my statement.

GnuBG code is public, I've looked at it (like anyone who make such claim should have) and there is no doubt that there is nothing that is done by the program incorrectly: it does not pick the dice, it does not look ahead the upcoming dice.

Conclusion: GnuBG does NOT cheat

Xavier Dufaure de Citres

To help get this thread back on track, I'd like to reiterate that my experiment was not meant to prove that gnu cheats. It was simply a test to find out if/why I play better when using manual dice. The error rates shown were not meant to prove anything, other than the fact that my play improves when I play with manual dice. I'm expanding my study to include individual checker/cube ERs, as well as luck ratings and match length.

I know the comments above were more directed at geneftw, but I just didn't want to turn this into a "my bot cheats" thread.

How do you get Gnu to record the moves?
Sorry. I misunderstood the intent of the OP. If I can get Gnu to record the moves, I'll start another thread comparing Gnu's dice vs manual, etc.

File save from gnubg will give you several formats you can save all the information in. But to have any significant evidence of gnu cheating as you claim you would need 10's if not 100's of thousands of data points and a fairly significant statistical analysis to go with it. The experiment I proposed is a simple way to demonstrate that the same seed produces the same rolls no matter what the position.

The only place I could find the moves listed were View>Panels>Game Record, but there's no option to print them and they cannot be copy/pasted. Oh well.

This isn't the first time Gene has gone off on GNU:

http://forumserver.twoplustwo.com/sh...55&postcount=5

Good luck with the more in depth study.

No, it isn't. That's the reason for my comment earlier in this thread, "I was tempted to post about it, but decided not to. Y'all would assume it's just another amatuer's 'Gnu cheats' thread", but when I gave NoChin's OP a glance, I got all excited and without reading carefully, I assumed it was a post presenting "Gnu cheats" evidence.

Sorry, folks. I didn't mean to hijack the thread. :blush:

There is a very important book that was published in the 1980s that supports your thesis. You may wish to acquire a copy of this right away!

http://www.bkgm.com/books/Ball-AlphaBackgammon.html

Once you finish a game, match or session just go to file save, you can save the file in several formats from there that will allow you go back and see all the rolls and plays.

Yes, but I can't print them.

I thank you, though.

EDIT:
I did try that File>Save thing, but I don't know what all those formats are. I tried one and when I opened it, a new game set-up was on my screen. I assumed that I could go View>Panels>Game Record from there, but I'd still be in the same boat. I was hoping to be able to copy/paste the results.

Save it in a plain text format. You can print and edit those as it's all written out in plain old text.

I just tried that. I saved a game as plain text and when I tried to open it, the "open" button was greyed out. I double-clicked it and got an error message, "Unable to import. Unrecognized file type."

This is crazy. I've saved four games, can't print anything from them and there's no way to delete them. My "remove" button is greyed out. Then I discovered I've got some games saved from some online play. I have no idea how they got there, but I can't delete them, either.... AAArrrrrgggghhhh!!!!!!!!!!!

Plain text format is not for gnubg to read. It's for human readability. So you need to just open it in a text editor like notepad (on PC) or textedit (on mac). Or you can always open it in MS Word, OpenOffice, or any other text editing program.