Re: f(x) and NN

[Nick Fortescue <nickf@xxxxxxxxxxxxxxx>]

On Fri, 7 May 1999, Joan Pons Semelis wrote:

> From: Weimin Xiao <ranxiao@xxxxxxxxxxxxxxxxx>
> > From: P.J.Leonard <P.J.Leonard@xxxxxxxxxxxxxxxxx>
> > ...
> > if moveSuggest() is perfect you do not need evaluate()
> > if  evaluate() is perfect you can suggest all possible moves and do a 1
> > ply search.
> > 
> > :- That's what I think only one function is enough, and evaluate() is more
> > fundamental. Only when evaluate() is not perfect enough, do we need few
> > plies to improve the result, or moveSuggest() to provide move candidates.
> > 
> > Weimin
> 
> for moveSuggest() to be able to work fine first some kind of evaluation
> of the board must be done, it will, at least implicitly, have to tell
> apart killable chains of stones from already dead and already alive; if
> every thing on board is dead or alive, and also if not, in fact, the
> moveSuggest() will have to find at least yose moves, and thit means
> that, again implicitly or explicitly, the previous work on the board
> can tell where borderlines of territories are, hence, provide the
> implicit (that tells moveSuggest() where to play) can be made explicit
> if it was'nt already, we can change this evaluation of the board in a
> good evaluate().
> If evaluate() is good enough, then it's easy to make a moveSuggest()
> that prune away most useless moves, and we don't need a deep ply search.
> 
> Bottom line: evaluate() *is* the fundamental.

This is not true. Consider the extreme case (which will never be achieved,
but we are talking theory, not practice here) of a move-suggest function
which just accesses a database (provided by God) of all possible board
positions, giving the best move for each. No evaluation need ever be done.
Of course, I am not suggesting this is practical (finding the database is
tricky) :-)

As P.J. Leonard wrote, either a perfect moveSuggest() or a perfect
evaluate() is enough.

Nick

--
Nick Fortescue                     E: nickf@xxxxxxxxxxxxxxxxx
Voice : +44 1483 448885            Fax: +44 1483 448845
Canon Research Centre Europe Ltd, 1 Occam Court, Occam Road,
Surrey Research Park, Guildford, Surrey. GU2 5YJ U.K.