Herein I foolishly attempt to determine whether BLAWN or BRAWN is really 
better,
and try to understand Quackle output.  At the end I put my Wish List so 
maybe
scroll to that if you are hurrying here.

Some possibly interesting info:

WORD COUNT

Beyond length 2 and 3,
there are more words with R,U in them than with N,U in them.

>From Zyzzyva searches

Length, RU* anagrams, NU* anagrams:
3, 16, 18
4, 136, 105
5, 471, 336
6, 1230, 955
7, 2519, 2051
8, 3683, 3209
Total, 24378, 23851
 (total is length 2-15)

However this doesn't consider the average score of the words, or 
probabilities - only counts them

VOLATILITY

BLAWN has no hooks or extensions
BRAWL has -ED, -ER, -ERS, -IE, -IER, -IEST, -ING, -S, -Y, OUT-, OUT-ED, 
OUT-ING, OUT-S

SIMULATION

By Quackle simulating 4 plies for 10,011 iterations there is just a tiny 
difference in how
good the plays are

but I want to look at what happens on each ply ...
8D BRAWL

  1.. (26 [sd 0]) (36.0856 [sd 24.7586])
  2.. (35.8721 [sd 22.0508]) (41.2835 [sd 23.9896])
  3.. (40.1245 [sd 23.4687])
  a.. Residual: -0.723356 (sd 12.8543)
  b.. Spread: 24.5932 (sd 43.8822)
  c.. Valuation: 23.9042
  d.. Bogowin%: 52.1581%
8D BLAWN

  1.. (26 [sd 0]) (34.9355 [sd 24.4112])
  2.. (34.0318 [sd 21.7761]) (39.6406 [sd 23.7178])
  3.. (39.0373 [sd 23.5788])
  a.. Residual: -0.525067 (sd 13.025)
  b.. Spread: 24.5902 (sd 43.4319)
  c.. Valuation: 23.9679
  d.. Bogowin%: 52.2026%
COMPARISON (leaving out the SD)
8D BLAWN (UR) - 8D BRAWL (UN)
1.  0   (-1.1503)
The two scores are the same (26)
With BLAWN on the board opponent scores 1.15 less
2.  -1.8403 (-1.6429)
With BLAWN and UR I score 1.8 less with my 2nd play after my opponent's 
reply
My opponent also scores 1.6 points less on his 2nd play after BLAWN
3. -1.0872
With BLAWN and UR I score 1.09 less on my third play
Conclusion:
For this simulation the bogowin and the evaluation tell me either play is 
equally good

3 ply simulation is also interesting.
The bogowin gives a different indication than the valuation.
CONCLUSION
Here would be my wish list for Quackle after this:
 * A listing of the top 10 words at each step in the simulation.  (Or I 
could log to a file and analyze it.)
 * A way to compare two plays side by side instead of having to subtract.
 * More detail on the 'bogowin'.  Now it's an estimate, right?  But what is 
it based on?  It seems like a histogram would be more enlightening for 
simulations in the beginning and middle game.

I could babble about the endgame but I don't know what it's currently doing 
so I'll hold back on that

Kevin Leeds.




Reply via email to