IA_evoman

Project developed to the WCCI competition, first introduced on UFABC's Artificial Intelligence class.

Files

• our_controller.py: Creates the Artificial Neural Network, preprocess inputs and sends the agent's actions to the game.
• our_optimization.py: Defines the environment and the Genetic Algorithm.

The outputs are presented in the our_tests folder, containing:

• evoman_logs.txt: logs from evoman generated during execution
• results.txt: logs generations and their best fitness
• Evoman.pkl: all ANN weights
• results.csv: the results for each iteration containing:
  • each train boss fight result and their average
  • each test boss fight result and their average
  • global average
  • gain (harmonic mean)
  • our custom fitness
• results.png: the graphic between the gain and the fitness function on each iteration

Environment set-up

Make sure you have python 3.6 or 3.7, pip and venv dependencies installed. If not, you can install them using:

Ubuntu/Debian:

sudo apt-get install python3
sudo apt-get install python3-pip
sudo apt-get install python3-venv

Then install the dependencies:

# Create a new virtual environment
python3 -m venv env

# Activate the environment
source env/bin/activate

# Upgrade pip version
pip install --upgrade pip

# Install all required dependencies
pip install -r requirements.txt

Running

Make sure that venv is active, then to run the program, execute python our_optimization.py

You can change between the train and test modes by editing the mode variable to 'train' or 'test'. It's also possible to select the enemies for the training phase editing the enemies variable in the same file.

On train mode, the graphic interface is deactivated and, at the end of iterations (defined on GA class), it runs a final test with the best agent.

Results and Conclusions

We present our results and conclusions in this paper.