Words With Friends Algorithm

[Limbic~Region]

All,
I have been rather distant from the Monastery for a year or more. My daughters are growing up and other hobbies consume more and more time. During a recent episode of The Big Bang Theory, I had an opportunity to be introduced to Words With Friends. This of course made me want to write an algorithm to play optimally which in turn made me think of the monks.

I am not interested in writing yet another cheater - there are plenty of them out there. I am interested in writing a platform where two computer algorithms can duel for bragging rights. There are plenty of strategies one might consider:

• Saving high value letters for premium squares
• Avoiding moves that open up premium squares to your opponent
• Considering which tiles remain
• Intentionally opening premium squares to your oponent which will in turn open even higher premium squares to you
• Determining the best place to pass or swap tiles
• Keeping a proper balance of vowels and consonants in your rack
• etc

If I built the platform, would anyone be interested in writing a player? If so, what ideas do you have? If you are not interested in writing a player, what strategies do you think would work best (so I can steal them for my player)?

Update: I should clarify. If you are interested in writing a player, I am soliciting ideas for how game play should be done. In other words, a web interface, being able to play the same game twice (taking turns with who goes first), etc.

[zentara]

Your posting is rather vague, but it reminds me of Scrabble. You might be interested in looking at ztk-babel client server game with chat for an idea of how to make a player. You can take an modify that code if you desire.

[eyepopslikeamosquito]

Perl monk thelenm, formerly an expert Perl golfer, is nowadays an expert Scrabble player and tournament director and author of Zyzzyva, a School Scrabble program. I emailed him just in case he has any advice on this node.

[thelenm]

[ColonelPanic]

This is a significant AI problem. The generic solution would be running simulated future game outcomes for each potential move to see which one maximizes your probable point advantage. However, this will get expensive very quickly--each move involves scanning thousands of possible word/location combinations, and your opponents' tiles are not known. It would get very interesting if you restrict the amount of time the player can run and the size of the code.

My suggestions:

• Think hard about a good set of rules. To keep this fun, you want to require the players to be fairly simple, and you don't want the games to take forever.
• To start out, just make a game module and post it here, along with a simplistic player that will give people a test opponent. Then people can download it and run it themselves with different players. If the idea draws interest, you can set up something more complex, like a web interface.

[ambrus]

This is to note that tom7's Scribble is a similar word game that also allows for similar strategy questions. That game is cooperative, not competitive, that is, players try to help get the highest shared score instead of defeat each other.