This project uses evolutionary algorithms to solve the function maximums search problem.
Algorithm properties:
- Parent selection - FUSS.
- Children selection: CROWD TOUR, ELITE, ALL, RAND, FUDS, and MOD FUDS.
- Mutation probabilities: 0.15, 0.75.
- Dimensions: 1-5, 10, 15, 20.
- Generation gap - N individuals.
The algorithm was tested on 4 test functions, and 15 real ones.
Each algorithm variation was run 10 times.
On the first run, we are creating a plot on each 300+th iteration (the exact number changes dynamically), and for other runs, only the resulting population is visualized.
For reach run, we also gather stats such as the number of seeds, real, false peaks, etc. in the form of Excel sheets.
To speed up the tests, they can be run in parallel. The script is zsh and Mac OS specific but can be relatively easily adapted to another system.
Run results are hosted on Mega.