Monday, September 13, 2010

Class Notes: 9/13/10

Remember to register for the course if you haven't.

Project 1 Introduction

Q/A:
Daniel: Are letters carried over between rounds?
KR: No
Zach: If rounds are independent, do the number of secret letters remain constant?
KR: Yes
Jatin: How are starting points associated with the number of secret letters?
KR: You always start with 100 points no matter how many secret/starting letters you have, but in each game everyone starts with the same number of letters.

KR: What do you think were some important ideas identified last year?
Elizabeth: Using a strategy that values letters based upon how others are bidding as well.
Jesse: Valuing a letter more if it kept paths to getting 7 letters high
Aniket: Integrating the history - remembering that if it was hard to get an A before, it might be hard to get an A again
Monica: Microstrategies are helpful in creating refinements vs. a monolithic strategy

KR: Remember that you have access to last year's players. This can be good, or it can be bad, depending on how they were written and what you plan to do. Remember that if you take last year's player, tweak a few parameters, and submit it as yours, you will not be evaluated as highly as a team who brings in new, fresh ideas. Taking someone else's code as a starting base is fine, but make sure to implement NEW functionality as well, which you can clearly attribute as your own work.

KR: What else was interesting from last year?
John: Preprocessing over the wordset (offline)
Flavio: You also need to have something fast online (ex, the apriori algorithm)
(JB: paper here, Ben's original code here, original comments from Ben: "The ScrabbleMiner file has a demo/test program that builds the data structure and then looks up the set of letters passed on the command line--the other two have the actual logic.  The rest of the code has fewer comments than it ought to, but isn't *entirely* incomprehensible, I hope.  If you want to comprehend it in detail, the attached paper would probably be helpful, though.")
Yang-ziah (sp?): Some players gave a bonus for getting a letter that would complete a 7 letter word (typically 15 points). Also important to track letter frequency (as it decays from the starting scrabble frequencies)

KR: Remember, I strongly prefer that you are not using laptops during the class, unless you are, ex, submitting your code so that JB can run it, or doing something very very necessary (JB: Don't bother taking notes at all except for maybe jotting a few reminders to yourself, just read mine).

Monica: Data management was super necessary, ex, removing anagrams from the dictionary
Eva: Using a loss-based strategy - sometimes you should bid high just to block somebody else from getting something
KR: That wasn't that useful last year actually. Could it be more useful this year?
Neha: Since we have an unlimited rack size, even if we won a letter we didn't want, we won't get killed by it
Pranjal: If we've already gotten our 7 letter word, bid to block others (or make them pay) instead of bidding 0
Zach: You might still end up winning and losing points
Lauren: If you're in a small group of players it might be useful still, or if you are ahead by a lot.
Dan: Will we know how many rounds we are playing?
KR: Yes - it's 8p
Dan: Actually, if you are behind, you might want to take some risks. KR - Or, have a "meta strategy" to say "am I doing well?" and if not, "do something different
DFed: You might be losing a lot to get these letters, because if someone really needs them, they'll bid high
Nitin: If there are no hidden letters, you have perfect information, and you can stop it. And you should. If there ARE hidden letters, then you can only approximate if you are blocking a 7 letter word.
Elba: Alternatively, have a micro-strategy that changes when anyone gets 7 letters
Archy: Is "winning" by total points or total wins? KR - Total points. Archy - So don't always win, just do above average
Jatin: Maybe you should just always bid high at the start, since you'll be the only one taking this risk
Blake: This could be very, very bad if someone else does the same!
Aditya: In the first round, if people re bidding a lot, then hold out and wait to see who does what.

KR: What are some important differences we should look at?
Andrew: Sniping may not be as effective this year, since there will be more and more letters. The premium to get the 7th letter may be lower
Daniel: Hard to do a lot in 1 second
Bo: Pay more attention to other people's bidding and create strategies based upon that
Yang-ziah (sp?): Should focus more on ourselves

KR: How big a deal is it that there are more letters possibly?
Neerja: Do all of the letters have the same # of points? KR - No, they are scrabble scored.
Archana: These extra letters force you to bid more points
Claire: Different strategies for different games (2/3/4 etc person games)

Deliverable for Wednesday:
Submit a player. It doesn't have to be great. It doesn't have to win. It shouldn't crash, and it should have something that is building towards a final strategy.

Please make sure to name your players in the package:
seven.f10.g<groupnumber>.<yourplayername>

No comments:

Post a Comment