Re: computer-go: Temporal Difference Learning

[Don Dailey <drd@xxxxxxx>]

   My conclusion and that of everyone i spoke with was ...

I think my results show that you need to rethink your conclusions.

Because I reported something much  better than what you descibe below.
If  your  results  were  really  this bad,  you  are  doing  something
different than I  am and you should be trying to  figure out what that
is.

   ... that within 1
   sequential movement through the evaluation function i can pick better
   values than TD learning in 2 years at a 500 processor machine will learn,
   under the assumption that the evaluation function is the evaluation
   function of a reasonable strength chessprogram with not too many bugs (and
   because of that first condition the number of parameters is several
   thousands trivially or more). What does it take me to go through the
   parameters of a chessprogram in 1 sequential movement, 3 hours or 6 hours
   at most for a big evaluation function?

   Setting up the TD learning experiment takes longer already...