opt

Alibrary for optimization methods written in C++17 with concepts.

This library provides a set of numerical optimization routines (for now, just genetic algorithms). Thanks to C++ concepts each routine is able to select adequate strategies and optimization methods according to the involved data types and the provided information.

Installation

This library is header-only (except for the examples). To obtain it you can use the command line:

git clone https://github.com/adolfomunoz/opt.git

To compile any code that depends on this library, however, you require both the C++17 standard and concepts activated:

g++ --std=c++17 -fconcepts -O3 yoursourcecode.cc

These are only available in gcc >= 7.0. This library has been tested with gcc 7.2.0. There are several examples provided, which you can compile from the command line as stated.

If you'd rather use cmake there is a CMakeLists.txt provided, which can be used as follows:

mkdir build
cd build
cmake -G "Unix Makefiles" .. -DCMAKE_BUILD_TYPE=Release
make

This downloads (or updates) all the required dependencies into the external subfolder and compiles all the provided examples and tests and moves the executable files to the bin subfolder. Other cmake generators (such as Visual Studio) have not been tested.

Documentation

This library tries to minimize any function you want through the minimize command, which has the following syntax:

opt::minimize(<function>, <method>, <optimization parameters>)

where

• <function> represents the function to minimize.
• <method> represents the method to use in the optimization.
• <optimization parameters> represents the per-minimization specific parameters (which vary depending on the optimization method).

Both <method> and <optimization parameters> can be omitted, and the minimize function will try to deduce the best optimization method (generally, genetic algorithms) together with the most adequate parameters, according to the data types involved in the fitting function:

int main(int argc, char** argv) {
	std::cout<<opt::minimize([=] (float x) { return (121.0f - x*x)*(121.0f - x*x); })<<std::endl;
}

• Fitting functions.
  • Passing function parameters
  • Fitting functions with multiple parameters.
• Optimization methods.
  • Genetic algorithms. (default)
  • Pattern search.

Dependencies

This library depends on callable.hpp for function traits.

Disclaimer

This code is released under the GPL-3.0 license. In practice, if you use this code I'd like the following to happen:

• Let me know that you are using it.
• Let me know how can it be improved.