Re: [gnugo-devel] Levels and time used.

[Gunnar Farnebäck]

Jens wrote:
> On my computer running as level 10 is realy fast. And now when the
> talk about specific patterns with high values for getting out of
> reading more quickly makes me think that that aproch does not realy
> 'improve' GNU Go.

Every speedup we can make while getting the same results is positive
and in a real sense an improvement (although not necessarily
worthwhile when complexity and maintainability of the changes are
considered). The reason for this is that it leaves us more space to
introduce other changes where spending more time does increase the
strength.

Whether GNU Go at the default level currently plays too fast is a
different question and a reasonable one. I would say that the main
resistance to slowing it down is related to the regressions. While
those are without question of crucial value for testing changes they
also do slow down the development in the sense that they simply take a
long time to run and would do so even more if the engine was slower.

Essentially there are two, non-exclusive, solutions to this problem.
1. Speed up the regressions.
  a) Remove questionable test cases.
  b) Simplify test cases where complex surroundings are not necessary
     to identify specific problems.
2. Automate and distribute regression runs in order to better utilize
   available computer resources.

Regarding the second option we could really use a system for automatic
testing and publication of results for changes going into CVS. In the
longer term it would also be extremely useful to have something
similar for proposed patches.

/Gunnar