> Trying to improve GLPK/GMPL 

It is a non-trivial task.

The MathProg translator included in glpk was implemented in a 
non-efficient way, because there was no intention to process models 
of huge size (so this allowed essentially reducing implementation
efforts).

The key idea to translate MathProg (or AMPL) models much more
efficiently is to avoid using "direct" representation of sets and
arrays and implement all operations in a way similar to one used 
in relational database management systems.

> I can see that GMPL model generation is 
> making an excessive number of set element copies (see bellow).

Please note that most of the copies might be temporary quantities 
that exist (allocated in the memory) only during evaluation of the
(sub)expressions involving these quantities.


Best regards,

Andrew Makhorin

Reply via email to