Show HN: Play rock paper and scissors against a untrained neural network

https://github.com/victorqribeiro/jokenpo

by atum47 • 7 years ago

101 63 7 years ago

github.com

thirdmeow• 7 years ago

Wow it's surprisingly good

r-w• 7 years ago

Or we’re surprisingly bad…

camjohnson263• 7 years ago

Check the other comments, looks like it's cheating by reading the human's input before guessing.

atum477• 7 years ago

no, it's not.

thdxr• 7 years ago

Hm in my limited testing of not looking at the screen and selecting as randomly as I could I was surprised to find it not end 50/50

Not enough trials or is there something else going on here?

quickthrower2• 7 years ago

2 things to consider...

1. Humans aren't good at random, try using a RNG or dice.

2. If you toss 10 coins, what is the probability of 5 heads and 5 tails? It isn't very likely, even though it's the most likely outcome.

mbrumlow9• 7 years ago

I let rand play it 500 times the results are as followed.

Player: 144, Computer: 179, Tie: 177,

Although I think I got more equal results when I did 1000 iterations, but failed to record that one.

https://youtu.be/g9Zo771HYpM

bobblywobbles• 7 years ago

I expected to lose against a computer.

I was right.

QuadmasterXLII2• 7 years ago

I hate to say it, but I’m not sure it’s playing fair. I used random.org to play truly randomly, and the “neural network” beat me to 10 pts 6 times in a row.

QuadmasterXLII2• 7 years ago

Looking at the source code, the move you are about to make is included in the training dataset. This can be confirmed by playing “scissors scissors scissors rock” and then looking at the variables x and y in the console, which will include the surprise rock.

The code updates x and y, then trains the model, and then makes a prediction. “Fair” code would make a prediction, then update x and y and train the model

This explains the behavior mentioned in the comments where the computer gets an impressive early lead due to the players next move being one of only a few datapoints it learns from, then backs off to a more plausible advantage as the leaked data is diluted by past data.

atum477• 7 years ago

after the player makes a move the nn makes ones and add it to it's training data. the move the player does after that is added as a counter move to the nn move. this way I treat the data as time series.

dTal• 7 years ago

In other words, the description is precisely inaccurate. Not only is it not untrained, it's trained against your exact plays through mysterious future-knowledge!

jchw2• 7 years ago

The thing is, if you are playing randomly, that makes it easier to predict what you’re (not) going to do next. A good random distribution would give fewer runs, which means your choice is less likely to be the last choice you picked and somewhat less likely to be other recent values. It can at least tie by choosing scissors if it thinks you will not pick rock.

This is kind of a shot in the dark because I am not familiar with RPS strategy, but I think it makes sense.

mbrumlow9• 7 years ago

But if it predicts you are playing random, is it not then random vs random?

I am under the impression that you can't game random -- unless you are aware of the random generator being used is broken..

EGreg• 7 years ago

That sounds like the gambler’s fallacy. Less runs than what? Most truly random input haa far more runs than what people “think” is random, and in fact that’s one of the statistical tests for whether a data set was random.

You’re essentially saying that a good neural network can predict the next value of a good random number generator. Good luck with that one!

Maybe while you’re at it, have neural networks invert cryptographically secure hash functions :)

seanhunter2• 7 years ago

Random specifically doesn't mean it's less likely to be the last choice you picked. At least if we're talking uniform random which we should be here. It means the last choice you picked is exactly as likely as any other choice.

wongarsu• 7 years ago

A good random distribution would give fewer runs

A good random distribution actually gives more and longer runs than a human trying to appear random.

mbrumlow9• 7 years ago

Yeah, I am not sure what is going on, I just played it 1000 times from rand and the results were.

500 plays - Player: 144, Computer: 179, Tie: 177

https://youtu.be/g9Zo771HYpM

1000 plays - Player: 325, computer: 359, tie: 316

https://youtu.be/7pB5TZ_xYzE

albertshin2• 7 years ago

How are you emulating the clicking?

camjohnson263• 7 years ago

I just did the same thing,

500 plays - Player: 163, Computer: 179, Tie: 158

1000 plays - Player: 306, Computer: 365, Tie: 329

I've run it several times and the computer has won every time against random, it must have an unfair advantage.

sagebird• 7 years ago

I wonder about an arms race between two advanced AI’s which play RPS.

I can’t fully put my head around this, but what would it be like if each AI could read the architecture of the other’s brain before each move. The AI’s would be permitted to reconfigure themselves as they play. An “obvious” strategy may be to simulate your opponent and ask what they are likely to play. Though, simulating their behavior is likely to involve you simulating someone else simulating your behavior, ad-infinium, until your computing resources bottom out. It is like fighting the man in the mirror who can choose to mirror you or not.

make3• 7 years ago

don't you just end up with a (discrete) uniform distribution on both sides? anything else can be exploited by the opponent

mikeash• 7 years ago

Yep, RPS is only interesting to the extent that the players are imperfect. It’s most interesting as an illustration of how bad people are at behaving randomly.

timfoden• 7 years ago

There's a site where AIs play RPS on an ongoing basis http://www.rpscontest.com/. Source code of the entrants is available. My best bot (called paper6) http://www.rpscontest.com/entry/5640686850277376 wins on average 67% of the time, but the top bots win around 80% of matches.

resist_futility• 7 years ago

Screw up the NN by selecting the same thing over and over at first. Then start alternating between two values choosing a new set of values when it starts guessing correctly again. Rinse. Repeat.

veselin• 7 years ago

I have "beaten" it. Play initially like you would - typically a human would change his choice often. The network would learn from you and gain a lead of say 10 points. After this, if you pick something that gives you a win start sticking to it while it gives you a win. You will quickly regain the lost points and gain a small lead. This seems to work every time I try it. After 60 moves I always lead a bit. Btw, the game gets slow then.

Not sure a bigram or trigram model would not do better than the neural network.

atum477• 7 years ago

I have merged the first pull request that changes to the prediction first training after.

sambroner• 7 years ago

Nice. Seems to beat me. I performed significantly better when I used random.org to generate numbers than when I played with my own attempt at randomness.

atum477• 7 years ago

now with a new interface, thanks to https://github.com/ocjojo

atum477• 7 years ago

This is my awstats after being featured here. Can someone tell how efficient is my code being served? I have no other numbers (or default values) to compare.

day | visits | pages | hits | bytes

26 Jul 2019 | 4,688 | 16,507 | 60,489 | 163.89 MB

Show HN: Play rock paper and scissors against a untrained neural network

Related Stories