Saturday, June 16, 2007
What have I been up to? -vvv

The past few weekends have been relatively uneventful. I had a cold for a while but most of the time, I was reading the book Cosmic Jackpot that Shuquan gave me. It was rather interesting and presented various points of view on why the universe seems fit for life but most of the theories presented were highly speculative. My inclinations are in a different direction from the author but without any substantial evidence, I guess it all boils down to personal preference. Perhaps I'll elaborate on this in a future blog entry.

I have also started a new personal project. I am writing a simple computer program that plays a simplified version of rock-paper-scissors with a human player. In my version of the game, the human chooses red or black and the computer does the same. If the colours match, the computer wins, otherwise the human wins. My objective is to create an algorithm to predict a human's next response based on his past responses.

So far I have tried out one method based on the randomness of the human player's response. It didn't perform too well, just barely better than random guessing. I have another idea which I will try soon.

I was gonna write more but it's about time I leave for the Play! A Video Game Symphony concert. Adios!

-- permalink --
Yiheng made 4:22 PM

2 Comments:

Curious to find out what kind of strategy you have assumed that the human is playing. If it is based on the assumption that players chooses black or white randomly (and that is in fact the best strategy to play if the opponent is playing his best strategy), then the computer cannot do any better than playing randomly.

It's already a Nash Equilibrium. However, unless you assume that the human plays in a particular fashion...it would interesting to find out what that is. To me, play play rock-paper-scissors randomly
By  Soqcrates, at 1:54 AM  

In my first approach, I assume the human will try to play randomly but fail at it. The assumption is that humans are generally poor at creating random sequences. For example, for a random sequence of 100 trials, it is likely to see 5 consecutive reds or blacks but few humans will play it that way. I am trying to see if I can capitalize on this non-randomness.

In my second approach, I assume the human will have a subconscious preference for a certain color depending on how he performed in the last few trials.

So far, my algorithms are not sufficiently adaptive. They take too long to learn the pattern and slow to pick up new ones, i.e. they keep sticking to old strategies even after the human player switches strategies. I think my program will only do well against children.

I think random is the best strategy only if your opponent plays random. Otherwise it would be better to use the opponent's predictability against him. For example, if the opponent is an idiot who always play paper, then the best strategy would be to always play scissors. Of course, in doing so, one becomes predictable. I imagine there will be some sort of decaying oscillation around the Nash Equilibrium with each player trying to outguess each other until they both become random.
By  Anonymous, at 8:17 PM  

Post a Comment

---