The incentive idea proved to be just a modest improvement and I tried
all kinds of weights and formula's.   The linear formula seems to work
best and gives roughly 30 ELO.   It definitely seems worth it.

However, I combined that with another improvement that gives a major
strength boost.  This new improvement is based on the observation that
one of the most common errors for these simple bots is self-atari.  The
most common case is to attack a big group, while putting yourself into
atari with the stone you are attacking with.   

The idea is trivial:  When scoring the moves at move selection time,
give a penalty for self-atari moves.   I am only considering
non-capturing self-atari moves.   I have only tried 2 values,  0.05 and
0.10 and 0.10 is working the best.

After over 350 games it is giving 131 ELO improvement over the standard
"mark williams enhanced" bot.   This bot also has the linear incentive
bonus in it.

This is well beyond the margin of error - so I can say with a great deal
of confidence that it is much stronger.   

- Don
  




On Thu, 2008-10-30 at 19:15 -0400, Jason House wrote:
> The error bars of all bots overlap. I'm not familiar enough with  
> BayesELO to compute p-values. I'd bet that only the 0.1 version has
> a  
> statistically significant strength difference.

Attachment: signature.asc
Description: This is a digitally signed message part

_______________________________________________
computer-go mailing list
[email protected]
http://www.computer-go.org/mailman/listinfo/computer-go/

Reply via email to