This program was created for the Ferienakademie course Modern Approaches to Optimization and Verification, where I worked on Particle Swarm optimization. The aim of this program was to visualize the Particle Swarm Optimization nicely, while still maintaining very good performance and proving the power of this method.
Because of the stated motivation above, I decided to split up the visualization to be implemented in Python and the PSO at its core in C.
The PSO-Module, written in C, is responsible for the computational work of the method. It takes references to the arrays of positions, velocities and past best positions of the particles and the global best position. With this information, it will operate on the array and update its values. The C program will be compiled into a shared library (look at the Makefile!), so that the visualization program can call it with the help of ctypes.
I used Panda3D, an open-source render engine, for rendering PSO as it promised to be fast and effective, as it is written in C++. It first takes initial inputs like the amount of particles and their restriction and then generates those initial positions and velocities, etc. Then it will pass those references to the PSO-Module and the values get updated. The results are visualized in a 3D-visualization.
- Use vectorization / intrinsics in PSO-Module to gain even more speed!
- Fix some visualization issues
- Implement Discrete Particle Optimization and its visualization
Those are some results of the current program.
Particle Swarm Optimization on Schwefel Function
Particle Swarm Optimization on Pizza Function