Browse files

Added readme.

  • Loading branch information...
1 parent ba46a86 commit cf2bdaf85e0f44f9636e5eba014245609471050a @amcnamara committed Feb 20, 2012
Showing with 17 additions and 0 deletions.
  1. +17 −0
@@ -0,0 +1,17 @@
+= Hangman =
+This is a basic predictive engine for [Hangman]( guessing strategies.
+== Analysis ==
+The strategy is based on the following formula, where:
+ t is the total number of words on the current branch
+ d is the number of possible character maskings that could result from the following guess
+ d_max is the maximum number of possible character maskings for all possible guesses on the current branch
+ n_i is the number of words that are a subset of a given sub-branch (character masking) from a given guess
+ c is a constant for fine-tuning the sub-group size weighting
+<img src="">
+The idea is to pick a guess character that will return the most possible information (based on returned character maskings, and the possible subset of words that each of those groups can eliminate). There are two ways to attack this problem: the first is to get as even a distribution as possible across the possible sub-maskings (for t=100 and d=4, n_(1...4)=25-25-25-25 is far more valuable than n_(1...4)=97-1-1-1), and second is to give preferrential weighting to groups which are more granular (higher d). The summation in the formula above will tend towards zero as the sub-groups tend towards an even and constant distribution, and the weighting on the right will tend towards c as the granularity increases; the guess for which this function is minimal will be the optimal pick.

0 comments on commit cf2bdaf

Please sign in to comment.