SpyOpt

SpyOpt is a C++ implementation of the metaheuristic spy_algorithm [1]. It can search for global optimal solutions for arbitrary non-convex functions.

SpyOpt はメタヒューリスティクスである spy_algorithm [1] の C++ 実装です。 任意の非凸関数に対し、大域最適解を探索することができます。

Demo

Ackley Function

Eggholder Function

Requirement

• A compiler that supports C++17
• yaml-cpp library

Installation

1. Install yaml-cpp:

   On Ubuntu:

   sudo apt update
   sudo apt install libyaml-cpp-dev
   

2. Clone the repository:

   git clone [email protected]:KentaKato/SpyOpt.git
   cd SpyOpt

3. Create a build directory:

   mkdir build
   cd build

4. Run CMake and build:

   cmake ..
   make

This will produce the executable file spyopt in the build directory.

Usage

1. Execute Optimization

   After building, you can run the spyopt executable:

   ./spyopt

   The results will be saved in results/history.yaml.

2. Animation of Otimization History

   cd SpyOpt
   python3 scripts/create_history_animation.py

   The gif file will be saved in current directory.

Configuration

How to Modify Search Parameters

Settings can be adjusted via resources/config.yaml.

num_agents:    50  # Number of agents. Should be greater than (num_high_rank + num_mid_rank)
num_high_rank: 10  # Number of high rank agents
num_mid_rank:  25  # Number of mid rank agents
num_iterations: 50 # Number of iterations
swing_factor: 1
objective_function: Ackley # Booth, Eggholder, Ackley

How to Customize the Objective Function

1. Implement it as following:

   double your_custom_function(const std::vector<double> &pos)
   {
       ...
       return ...;
   }

2. Set the search space in resources/config.yaml.

   lower_bounds: [-5., -5.]
   upper_bounds: [5., 5.]

3. Pass the custom function as an argument to the SpyOpt`` constructor and execute the optimize()` method.

   Minimum Code Example:

   #include <iostream>
   #include "SpyOpt/spy_opt.h"
   #include "SpyOpt/config_parser.h"
   #include "SpyOpt/objective_functions.h" // Assuming you've added your custom function in this header
   
   int main() {
       // Define a simple custom objective function
       auto custom_function = [](const std::vector<double> &pos) -> double
       {
           return pos[0] * pos[0] + pos[1] * pos[1];
       };
   
       Config config;
       if (!parseConfig("resources/config.yaml", config))
       {
           std::cerr << "[Error] Failed to parse config!" << std::endl;
           return -1;
       }
   
       SpyOpt spy_alg(config, custom_function);
       spy_alg.optimize();
   
       const auto [fitness, pos] = spy_alg.getBestFitness();
       std::cout << "Best solution:" << std::endl;
       spy_alg.printBestAgent();
       spy_alg.dumpHistory("results/history.csv");
   
       return 0;
   }

Reference

[1] Pambudi, Dhidhi, and Masaki Kawamura. "Novel metaheuristic: spy algorithm." IEICE TRANSACTIONS on Information and Systems 105.2 (2022): 309-319.