I would like to toss out an idea to improve opening books for computer
programs.

In the KvXF match game 3 it is clear that K took advantage of a
three-fold failure in the opening book of XF. Namely the engine
allowed a quite opening rather than playing to it strengths (IE
sharper openings). That the opening it had chosen was not only
quiet, it was refuted with Qa4 that was not in it's book, and finally
that XF was stuck without a plan out of the opening.

It is point 3 I would like to address.

Openings tend to lead to thematic games. But computers tend to treat
all positions as equal, without understanding. They're eval functions
remain static regardless of the situation.

The general eval functions, tend to function pretty well across a
variety of situations, and the computers will often find and follow
thematic schemes, because they are the obvious best choice within
their event horizon, however there are all types of openings where it
is *not* obvious to the computer, and in these situations computers
often flail badly with no idea of what to do, even though it is
obvious to it's human opponent through experience what the theme and
methodology going forward should be...

I would suggest the following...
Out of opening "hints". This would describe eval tunings that would
be specific for this opening. This would give greater weight to the
thematic properties of the opening, and in some sense provide a "plan"
for the computer. A secondary type of hint would be a "thematic"
move, this may be an eventual sacrifice, piece placement or pawn
structure etc... In this case you would allow the engine to make move
extensions down lines that include the thematic positions and when
doing move sorting place thematic positions at the top of the list
giving more time for them. Again this would provide a computer a
"plan" and the computer would be able to tell when it would be most
appropriate to begin executing on that plan...

Any thoughts from folks???

Cheers

On Fri, 21 Nov 2003 10:13:42 -0800, Fortknight wrote:

I would like to toss out an idea to improve opening books for computer
programs.

In the KvXF match game 3 it is clear that K took advantage of a
three-fold failure in the opening book of XF. Namely the engine
allowed a quite opening rather than playing to it strengths (IE
sharper openings). That the opening it had chosen was not only
quiet, it was refuted with Qa4 that was not in it's book, and finally
that XF was stuck without a plan out of the opening.

It is point 3 I would like to address.

Openings tend to lead to thematic games. But computers tend to treat
all positions as equal, without understanding. They're eval functions
remain static regardless of the situation.

The general eval functions, tend to function pretty well across a
variety of situations, and the computers will often find and follow
thematic schemes, because they are the obvious best choice within
their event horizon, however there are all types of openings where it
is *not* obvious to the computer, and in these situations computers
often flail badly with no idea of what to do, even though it is
obvious to it's human opponent through experience what the theme and
methodology going forward should be...

I would suggest the following...
Out of opening "hints". This would describe eval tunings that would
be specific for this opening. This would give greater weight to the
thematic properties of the opening, and in some sense provide a "plan"
for the computer. A secondary type of hint would be a "thematic"
move, this may be an eventual sacrifice, piece placement or pawn
structure etc... In this case you would allow the engine to make move
extensions down lines that include the thematic positions and when
doing move sorting place thematic positions at the top of the list
giving more time for them. Again this would provide a computer a
"plan" and the computer would be able to tell when it would be most
appropriate to begin executing on that plan...

Any thoughts from folks???

Cheers

I believe attack tables/piece tables do just this. Give the program a
heads up in the eval for thematic continuations from the openings.

Such tables have to decrease in importance as part of the eval as the
board moves away from the end of book positions, and into untested
positions.

darrz