[Coin-discuss] Multicriteria optimization

[Emilia Tantar]

First, I thank you for your time and for your answer
and I also must excuse myself for the lack of details,
all in all hoping that I will be able to answer as
best as possible to your questions. 

A separation is made between the algorithms and
problem data representation model by using interfaces,
eliminating the Pandora's box problem. 

Algorithms access model data by making use of
interface functions assuring uniformity and a high
degree of abstractization. The user has to implement a
minimal set of functions for describing specific model
aspects, for the case of imperative model, allowing in
this manner the use of a larger set of representation
models. 

The imperative model is conceived having as purpose
data representation for genetic algorithms,
metaheuristics or representation models for
multi-criteria algorithms that do not use linear or 
nonlinear representation models.

We designed the model in order to achieve simplicity,
offering the possibility for the user to choose a
method from an underlying layer of algorithms as well
as offering the possibility of specifying a custom
algorithm for specific parts or for the entire
process.

The user can specify data using a linear model, a
non-linear model or an imperative manner, thus
offering a larger area of problem representation and
an easier way of integrating differrent classes of
algorithms in the COIN-OR infrastructure.

Unfortunately for the moment there isn't an online
version for the software we developed as for the
moment we are hoping to finish completely the initial
phase in order to make sure that the idea we followed
is viable.

Caught by the stream, I overpassed going deeper in
details regarding the mentioned "imperative model" -
you were completely right, adding in our vision a few
other things to our goal. 

All the way, this is accomplished by requesting as low
as possible data from user part, providing a default
resolution pattern for specific directions, mantaining
in the same time the flexibility by allowing complete
re-specification of the resolution pattern.

Interfaces are used for clear delimitation of
algorithms and data representation - in this manner
the algorithm makes use only of interface functions,
ignoring completely the model of representation. All
these keeping in mind that a complete unifcation is
not completely achievable and unnecessary for reasons
of complexity.


Thank you in advance for your responses!

Sincerely yours,
     Emilia Tantar


	
		
__________________________________ 
Do you Yahoo!? 
Yahoo! Mail - You care about security. So do we. 
http://promotions.yahoo.com/new_mail