This repository contains a project realized as part of the Evolutionary Computing course of the Master's degree in Artificial Intelligence, Vrije Universiteit Amsterdam.
The aim of this project is compare two Evolutionary Algorithms for the task of video game playing using a python framework called EvoMan. For more details, read the task assigment. The proposed solution is describer in the report.

For the first part of the experiment, meaning the evolution part, you need to:

• Run the approach1/genetic_optimization.py file. There you can set up the ENEMY global variable to change through different enemies.
• The best individual found is stored in the approach1/runs/enemy_#/best_individual_run_#.txt file. The history of the evolution is stored in the approach1/runs/enemy_#/logbook_run_#.csv file.
• The hyperparameter_tuning.py file is used to tune the hyperparameters of the algorithm through hyperopt
• approach1/experiment_runner.py runs 10 optimizations simultaneously.

For the second part, where the best individuals are confronted to different enemies, you need to:

• Run the approach1/play_with_best.py file, selecting the enemy you want to play with.
• Results are then stored in the approach1/runs/enemy_#/games_played.csv file.

For the first part of the experiment, meaning the evolution part, you need to:

• Run the approach2/neat_optmization.py file. There you can set up the ENEMIES global variable to change through different enemies.
• The best individual found is stored in the approach2/runs/enemy_#/best_individual_run_#.txt file. The history of the evolution is stored in the approach2/runs/enemy_#/logbook_run_#.csv file.

For the second part, where the best individuals are confronted to different enemies, you need to:

• Run the approach2/play_with_best.py file. There can change the ENEMY parameter to face different enemies (those enemies need to have a created folder and results from the first part of the experiment)
• Results are stored in the approach2/runs/enemy_#/games_played.csv file.

Results obtained with both approach show that even in the early generations we can obtain very good individuals, reaching fitness values above 90 for all three enemies tested. To have a deeper insight on the graphs, check out the plots folder.

This project is licensed under the GNU General Public Licens - see the LICENSE file for details