Vizsage

This isn't exactly within the ambit of this blog but at least it's about data. While I should have been doing work, I instead made an awesome spreadsheet to find rules of thumb for what the best Scrabble rack leaves are. (rules of thumb below, tables here: http://vizsage.com/other/scrabble/RackLeaveRules.html) The computer program Quackle is one of the strongest scrabble players in the world. It uses the following 'Superleave' valuation:

• http://web.mit.edu/~jasonkb/Public/scrabble/superleaves
  (warning: huge-assed file)
• http://vizsage.com/other/data/superleaves.xls
  (The above, sorted by value, only up-to-4 leaves)

To find "synergies" and "anti-synergies" (dysphoria?), I calculated the marginal valuation for each combo. Basically, how much of the value for each two-leave is explained by the valuation of the component one-leaves, etc? For example,
- From S (7.35) and M (0.08), the joint valuation of MS is 7.44, a marginal gain of 0.1: the joint valuation is almost entirely from S&M. (<-- will lead to interesting google hits). This combination has no synergy.
- From Q (-9.0) and U (-5.1), the joint valuation of QU is 0.2, a marginal gain of 14.3. This is by far the largest synergy; next is ZO at +3.2.

I also played with three-letter synergies -- 3-leave valuations marginally different from the most explanatory 2-leave. General Lessons:

• Get a feel for the 1-leave list, and the learn these:
  • Synergy: QU OZ JU CH GN WY IN DE JK ER EV GIN JKY JKU ERS KWY HWY ?IN EST JOW ?AL ?EL ?IL IST
  • Anti-Synergy: BP CG FP MV PV CW CQ QS SX LQ BV SZ QR BC CZ VZ MQ RX GQ + most things with blank BPV CGQ BCG LQR FPV LNQ SVZ CMQ CLQ BCV BNV KTV LMQ GKT CFV GMQ FSV LNR DGT
• Worth keeping with a blank: The letters in "Lei an orc DTM" + the following digrams IN AL IL EL CI AN ER EN AC AR IT NO QU ET DE CO AT OR LO GN OT AM DI CE IM IR DO MO GI AB AG
• double letters are bad (duh), except FF, which is good.

See the spreadsheet at http://vizsage.com/other/data/superleaves.xls). Don't go betting the house on these results....

Tables (including 1-tile-leave values) are available here.

Labels: data, nerd nerd nerd nerd nerd, scrabble

Posted by flip on Friday, September 7, 2007 at 9/07/2007 04:02:00 PM | Permalink