Rock-Paper-Scissors

Andreas Junghanns writes: "Check out the Second International RoShamBo Programming Competition for a completely different experience! If you think you know everything about Rock-Paper-Scissors -- here is your chance to prove it against some stiff international competition. At the Web site you can find rules, sample programs and a report of the first contest, complete with results and program descriptions." This looks pretty cool, and it might make a neat first project for someone, too.

It is....

From someone who has spent about 1 week of his life trying to come up with a startegy to win, I can tell you it is quite difficult. My strategy was to have three variables, each corresponding to the three possible moves - rock, paper or scissor. Then follow the rules below to come up with a move.

• Choose a random move. Add one to that move.
• Check if opponent is making same move all the time. Add one to the move that will beat that.
• Check if opponent is moving in a cyclic pattern. Add one to the move that will beat that.
• Check if opponent is trying to beat my last move. Add one...
• Check if opponent is trying to beat the move that I would have played if I wanted to beat it the last turn. Add one...
• If I've been losing a lot lately, make a random move, otherwise make the move that has the highest count.

These strategies all together easily beat the sample bots but get hammered by the competitive bots. They are no where near as powerful as the competitive bots. I was going to do a bit more extensive pattern matching routines to try and beat them, but just havent got around to doing it.

Vivek Mittal Telstra Research Labs

PS: I wonder how much my first post ever on /. will get moderated. :)

PPS: Can we add another way to get your account information back... enter email... I dont remember by user id and when I try to create a new account, /. complains of a duplicate email.

I won the last year's competition -- here's how...

My submission, Iocaine Powder [ofb.net], won last year's competition. Follow the link to see a complete description of how it works. The competition results [ualberta.ca] from last year describe some of the other strategies that did well (and some that did not-so-well).

This competition is more complex than it seems; not only are there deliberate "dumb robots", but many of the real entries are quite predictable. A random player wouldn't have made it close to winning, and stalemates were rare.

What does this year hold in store? We'll just have to see!

Typical of Slashdot to glorify violence

I'm really not suprised that all you gun-toting, neo-nazi Americans would try to glorify some excessively violent childhood game like Rock Paper Scissors. Have any of you stopped to consider what these sorts of values these pasttimes instill in our children?

I mean, let's start with the rock. And I'm not refering to that movie with Nicholas Cage & Sean Connery in it, either. Rocks == Violence! Ask any caveman! Were it not for Oog being silenced by the Lameness Filter, I assure you he would back me up on this.

As for the scissors, well why don't you just throw children off a cliff? How many times have we been told not to run with scissors, and here /. is urging people to use them as both toys and weapons!

And the paper... oh Lord, how irresponsible can you get? We do all we can to squash that horrible "Puff the Magic Dragon" degenerate druggy song and then you people come along and start handing out Zig Zag's to elementary school students!

While we're at it, let's review the "premise" of this whole "game":

• Rocks violently destroy scissors
• Scissors violently slice apart paper
• Paper violently smothers the helpless rock

Are any of you thinking about the children? I seriously doubt it!

Two things

1) AI class I took last year (well, TA'd) we went in to some detail on possible RPS strategies. Yes, you can have strategies, at least assuming that your opponent is not a pure RNG (in which case the only correct strategy is to be RNG yourself.)
2) Nothing that the tourney produces will be as cool as this [duke.edu]. Unfortunately, the picture stinks, but on the left is my professor, and on the right is the kid (he'll hate me for that) who build the RPS-playing Lego Mindstorm. And that's the RPS bot in the kid's hand. It used some pattern learning software (written in legOS [sourceforge.net]) to attempt to detect patterns in human RPS players. Didn't work great, but what the heck... it was still damn cool. Had fingers and the whole bit.
~luge

Of RoShamBo and the Princess Bride

It's just like that game in the movie "Princess Bride", where a man reasons that to poison your drinking partner, you put the poison into your own glass. If he is suspicious he will switch with you when your back is turned, and will bring about his own doom. However, if he is a step cleverer than that, it becomes impossible to outsmart him - as the reverse-reverse psychologies pile up, the game boils down to random chance. Of course, the hero in the movie knew this and poisoned both glasses. After his opponent got thorougly lost in a maze of pseudo-logic, he took his poison and that was that.

That's the kind of visual image I get of someone trying to write a program that would win this contest - the "inconceivable!" guy from princess bride.

I wonder how a simple markov chain would do. That's where the probability of every move is based on the outcome of the previous game. For instance, "2 of the 3 times his rock beat my paper, his next move was scissors, so since his rock just beat my paper again, I'll anticipate scissors this time and go rock." I think this kind of reasoning would beat your typical human roshambo player in the long run, since a human would typically have a certain response based on what just happened.

Obviously, it's different with a computer. The program might anticipate this kind of thing, and has no general "feeling" that would you any reason to link a round to the one that came before. The more I think about this, the more I think it's just a matter of guessing right what other people will do.

--
grappler

Personally I prefer the South Park roshambo

Hey, it could be fun. make a quake bot that aims for the crotch, and dodges crotch shots. now that would be great for the AI people. Wait, that would be a great mod. Someone make NutShot Quake.

Play Rock, Paper, Scissors over the phone

Hey, check out:

http://studio.tellme.c om/home/documentation/example-111.html [tellme.com]

It's a company that produces a "VXML" platform that let's you program a phone voice system. Sample code #111 is a rock-paper-scissors game. Basically, you call up and play against a whiny, simulated kid voice. You can even "say" your commands...

In order to view the source, etc. you need to get a free login of their "developer studio" - but if you don't want to do that, here's how to play:

1. Call Tellme Studio: (1-877-461-3597)
2. Enter T-R-Y-I-T (87948) as the Developer ID
3. Enter T-R-Y-I (8794) as the Pin
4. Enter the code example's 3-digit Code Reference ID (in this case, 111)

Enjoy!

self-fulfilling?

Q: Can I enter Random (Optimal)?
A: No. You shouldn't want to anyway, because it is guaranteed to finish in the middle of the pack. It definitely will not finish in first place, because it cannot exploit the weaker programs.

This is a self-fulfilling prophecy isn't it? The more people believe the above statement, the more there is to gain or lose from a non (uniform) random strategy.

No, this is not a real tournament.

In a real tournament, you don't factor in just how badly you kick the asses of the worst players. This makes it a silly game of "guess who's playing". 2nd order roshambo: as silly as 1st order roshambo, and as painful as roshambeau.

For example, say there is one program that always uses rock (GOR), one with a sophisticated adaptive mechanism (IP), and 10,000 that always use scissors.

IP does very well, against these trivial opponents: on average missing the first 1, then recognising the pattern and getting every one after that.

GOR, however, wins every match, except against IP, against whom it loses every match after the first one (which is a toss-up).

At 1000 points per match, this gives GOR a score of 10 million and IP a score of around 9 thousand less than GOR. GOR wins over IP, despite the fact that IP beat every opponent GOR did, and beat GOR to boot.

It's all about the opponents. Remember that they're fluffing it up with enough ultra-stupid dummies that you don't have a hope in hell unless you beat these dummies soundly. Without the dummies, and in a real elimination tournament, a random-bot would have a fair crack at it, and there wouldn't be any point to it.

Of course, random(optimal) could still win, just as a monkey at a typewriter could recreate the complete works of Shakespeare, but it would take countless billions of tries for one to beat this system; there aren't enough people with computers to beat it that way. Of course, since the bots are the environment, if several dozen random bots were to enter for every non-random bot (including dummies), they would wash out the results in random noise (for every dummy you beat by 600 points, you'd face 50 random-bots that would randomly change your score up or down by, say, around 100 points), and all programs would be about equally likely to win (so a random-bot would probably take home the prize).

They have to restrict random-bots or strategy could become irrelevant and it would be 1st order silly, instead of 2nd order silly 8P

(and it is 2nd order silly; the basic way to win is simple to state, if complicated to implement: beat the trivial stupids, beat what you made to beat the trivial stupids, then beat what you have now, but the more levels of trickery you detect and beat, the more guesses you waste screwing around figuring out your opponent's strategy and the smaller the margin of victory, so you have to prioritize what level of trickery to try first, ending in the same sort of random guess that characterizes roshambo, except that you've spent a lot more effort...)

Cheater bots

Several cheater bots were entered in the last tournament. They were disqualified, of course, but here are the funniest ones:

• Fork Bot: Every move, this bot would fork itself into 3 processes and make a different move in each one. Any process that lost would be killed off in the next round, with the winning process continuing the tournament. Thus, you would think that it would never lose. However, when playing against the Psychic Friends Network, all three moves resulted in a loss, causing the Fork Bot to kill itself off, ending in a forfeit.
• The Psychic Friends Network: This program won 998/1000 rounds against any opponent other than The Matrix. No one really knows how it works, being incredibly obfusicated, but it appears to mess with the stack directly, among other things.
• The Matrix: Based on the simple premise "There is no spoon", this program won every single round of every match it was in. (Being written by the author of the tournament software, this was not very hard)

For more info, see this page [ualberta.ca] (near the bottom).

------

Re:This Must Be More Complex Than It Sounds . . .

The catch is that they put in some deliberately dumb robots, so that if you just use the optimal mixed strategy (randomness), you've got a 50% chance of beating them, but you can clobber them if you've got an actual plan.

In their FAQ, they tell you not to submit the random strategy, because it'll be guaranteed to finish in the middle of the pack.
--
No more e-mail address game - see my user info. Time for revenge.

Random number generators

From the FAQ (bold is mine):

Q: Can I produce my own random numbers, or must I use random(), and the provided flip_biased_coin() and biased_roshambo()? >/p>

A: You may use your own random number generator, but it must use a fixed seed, so that the tournament results are reproducible, given a fixed seed to srandom(). However, there is little to be gained from using your own RNG.

It seems to me that since they've also told you that random() is to be used, someone very clever could try to predict the opponent's choices based on sequence random() is returning. You aren't allowed to reseed it of course, but if your code is getting a certain sequence of numbers, is it possible to write code to figure out the current seed, and thus the entire sequence of numbers? Based on where your code winds up picking up the sequence, you know how many random numbers the opponent generated each round. Using that, you can possibly draw a correlation between the numbers you know he's getting and the choices he makes.

Granted, this is a longshot, and I know I'm not that clever, but on the other hand, there are lots of random number generators out there free for the taking. I'd spend the few minutes to add one to my code just to guarantee an attack like this won't work.

The real contest:

Who can anticipate the greatest number of lousy implementations that will show up?

You really have to hard-code recognition of the basic categories, since you aren't allowed to take the time to do a thorough analysis.

So, forget clever coding. It's grinding through all the bad ideas that will win this one.

Rock Paper Scisors expansion kits...

So, at one point, I knew this woman who thought rock, paper, scissors was sorta fun, and every so often, she'd insist on playing it. Her boyfriend got sick of this, and he would just point a single finger and say "Gun. Gun beats everything." So, once, she was playing me, and I was winning, and she tried this.

The next round, I did a sprawling-hand-spider. She said "gun." I said "space alien. Space alien is immune to gun", and I won the round.

Eventually, we also added dynamite and little bunny foo-foo, and rules for interactions between all the things. Everything beat three things, and lost to three things, so it was still balanced.

I don't remember all of the interactions, but the ones I do remember are funny.

"Townspeople throw rocks at alien" (rock beats alien)
"Little bunny foo-foo picks up alien and smacks it on the head". (lbff beats alien.)

The game has room for infinite complexity if your meal hasn't arrived yet.