Re: [Cilib-users] Calling cilib from existing code

[Edwin P. <ed...@pe...>]

On Sunday 18 May 2008 15:59:30 David Rooke wrote:

> Also, will CILib bark if, in my custom OptimizationProblem, I leave the
> dataset empty? As I plan to implement this, my fitness function will
> understand the data domain and feedback to the algorithm based on it's
> search of the parameter space. The way CILib seems to be written it expects
> the user to pass the test data into it.

Hi David,

You do not need to pass your test data to CILib. The framework will pass  
candidate solutions to 
net.sourceforge.cilib.problem.OptimisationProblem.getFitness(), wherein you 
are responsible for calculating the Fitness in whatever way is most 
appropriate for your problem. Thus, only your OptimisationProblem 
specialisation need have access to your dataset and know how to measure the 
candidate solution against your data.

As far as the search space is concerned, the framework will determine the 
domain by calling the OptimisationProblem.getDomain() method, wherein you 
must construct an appropriate DomainRegistry that describes your problem 
space. 

The DomainParser can be used to convert a string description of your search 
space (as described by Gary in a previous email) into an appropriate built 
repsentation (ie. a net.sourcefoge.cilib.type.types.Type representation). 
Refer to net.sourceforge.cilib.functions.Function for an example of this. Of 
course, there is nothing preventing you building your own Type representation 
either, if it's easier to map that onto your internal representation instead.

Hope that helps...

Gary, just thinking aloud here... I don't think the setDomain() method in 
Function is the most appropriate place to parse domain strings. Shouldn't 
this logic rather be pushed down into a specialisation of DomainRegistry, say 
StringBasedDomainRegistry which accepts the domain string in its constructor?

Then, the domain registry interface need only expose the 
getBuiltRepresentation() method which is used by the framework, and Funtion 
can return a new StringBasedDomainRegistry instead. This would encapsulate 
domain building from strings more cleanly in its own class and enable people 
like David to define his own mapping from problems to an appropriate "built 
representation" more intuitively as another subclass of DomainRegistry.

Regards,
Edwin Peer

-- 
#include http://www.peer.co.za/~espeer/disclaimer.html