Glen wrote:
> No.  Adjusting a rule is entirely different from adjusting a number.
> The adjustment of a number merely explores a space.  A number spectrum
> does specify/describe a metric.  So, for example, adjusting an integer
> with particular boundaries for the model, say [-10, 100] provides a
> well-defined space.
For a fixed instruction set there's a fixed set of programs that can be 
encoded in a fixed sized vector.   The behaviors that such a program can 
exhibit are also entirely fixed given precise initial state.  General 
and effective methods for global search can in fact be exactly the same 
for numbers and rules:   0) create a set of starting candidates 1) 
evaluate them, 2) tweak the good 3) destroy the bad, 4) go to 1.

To have good optimizations for searching  number spaces (more efficient 
than exhaustive grid search), then additional assumptions need to be 
made, such as that the numbers come from a differentiable function or 
have systematic gradients.   For that matter [-10, 100] is not a well 
defined space for a model because there are no units, and no given 
meaning to how that range ought to relate to sensitivities in other agents.

An agent model is an assembly.  If a component of the assembly is 
tweaked a bit, that doesn't justify calling it a whole new model any 
more than if a few parameters in the model changed a bit.   It is a 
versioning issue.

Marcus

============================================================
FRIAM Applied Complexity Group listserv
Meets Fridays 9a-11:30 at cafe at St. John's College
lectures, archives, unsubscribe, maps at http://www.friam.org

Reply via email to