Re: Advice on evaluation

[Heikki Levanto <heikki@xxxxxx>]

David Fotland (fotland@xxxxxxxxxxxxxxxxx) wrote in lsd.compgo:

: I'm thinking about evaluating differently based on who is moving next.
[...]
: So I don't want to do it if it is not going to help much.

I have never got that far with (any of my) programs, so I can not give
really tested advice. But I can't help thinking that it might be possible to
test your theory without making the large changes in your program. I have to
admit I am shooting in the dark, not knowing your program, but would
something like this work:

When ever evaluating a group, look half ply (one move) ahead with your usual
routines, and evaluate the group again (with your old evaluator). Now you
have two numbers, which you can log, or decide to take the better or worst
of them, or whatever. My guess is that a little systematical playing with
these numbers might give you a good feeling about the value of this
two-number system, without doing any radical surgery in your program.

Maybe you can implement several variations at the same time, play a hundred
games, and note where different ways would have played differently...

To me it sounds (almost) obvious that a group can have two values, depending
who moves next. "Black to live - White to kill" is not uncommon problem
type, especially for beginners. 

But it sounds (almost) as obvious that at some point you will have to
condense the values into one number, like the value of a proposed move, so
you can choose one.

Difficul questions! Thank you for sharing them!

	Heikki Levanto

--
Heikki Levanto  LSD - Levanto Software Development   <heikki@xxxxxxxxxxxxxxxxx>