Generalized PROgram for OPTimization
GPROPT is a program coded in Fortran intended to provide a generic tool to minimize a user-defined objective function. Several classical methods are implemented along with a genetic algorithm.
This program has been coded initially in 2015, in the framework of an assignment for a doctoral program. I have decided now to make it available under the MIT license. My expectation is that it becomes useful to anyone else. Contributions to improve it are also welcomed. A pull-request system is required to merge for the master branch.
The report of the assignment has been included in the repository: doc/Igor_OPT.pdf.
There you can find some theoretical background of the methods implemented, an overview of the program, and application examples.
A Makefile has been included to compile the code with Intel Fortran compiler ifort.
It has been tested in Ubuntu OS.
It is ready to compile with other compilers and in other OS, but some adjustments will be needed.
I hope to fix these issues in following versions.
You just need to open a terminal in the directory src/ and type make.
To compile it in release mode (optimized) just type make optm.
Objects are stored in a directory obj/ and the executables will appear in bin/.
Alternatively you can use any IDE suitable for Fortran (e.g. create a Fortran project in Visual Studio).
When running the executable GPROPT(_debug), the number of design variables and the optimization method are prompted.
For uni-dimensional problems the following options are available:
Uni-dimensional problem
Select Optimization algorithm:
1 - Polynomial approximation
2 - Golden Section Method
3 - Genetic Algorithm
0 - QUIT
For a general multi-dimensional problem, the following prompt is displayed:
Select Optimization algorithm:
1 - Powell Method
2 - Steepest Descent
3 - Fletcher-Reeves Conjugate Direction
4 - Polak-Ribiere Conjugate Direction
5 - Quasi-Newton Methods (Variable Metrics)
6 - Newton Method
7 - Genetic Algorithm
0 - QUIT
Additional options will be displayed according to the method selected.
The output is written for the terminal a an results file <method>.res.
The user must define an appropriate interface to evaluate the objective function value in src/evalfunc.f90
(or src/evalfunc1.f90 for the golden section method).
Some examples of objective functions are available in this routine.
The results of the examples addressed in the assignment report are available in examples/.
Send me an e-mail to [email protected] if you have any question or comment.