Evolutionary algorithm project

This repository contains code for running and analyzing two different evolutionary algorithms (EAs) for a multi-objective optimization task (designing a generalist controller in the Evoman framework).

Correspondence with report

• EA1 corresponds to Baseline EA in the report.
• EA2 corresponds to Neuroevolution of Augmenting Topologies algorithm (NEAT) in the report.

Algorithm Scripts

The scripts that do the heavy lifting and contain the class implemeting each algorithm, called by other scripts:

• In EA1_files: EA1_optimizer.py contains the code for the Baseline EA class

• In EA2_files: EA2_NEAT_optimizer.py contains the code for the NEAT EA class

Running the algorithms

To run the evolutionary algorithm once, execute the following scripts, with the appropriate command-line arguments:

• In EA1_files: main_EA1_single.py for EA1 (Baseline EA)

• In EA2_files: main_EA2_single.py for EA2 (NEAT)

Running multiple RUNS

To run the optimization n times for any enemy set and create the corresponding report folders, use:

• In EA1_files: main_EA1_multiple.py for EA1 (Baseline EA)

• In EA2_files: main_EA2_multiple.py for EA2 (NEAT)

Bayesian Parameter optimization

The following scripts perform parameter optimizations for the algorithms:

• In EA1_files:bayesian_EA1.py for EA1 (Baseline EA)

• In EA2_files: bayesian_EA2.py for EA2 (NEAT)

Data retrieval and preparation

To retrieve the data from the report results and prepare the gain data for boxplots, use:

• In EA1_files: get_gain_testing_phase_EA1.py for EA1 (Baseline EA)

• In EA2_files: get_gain_testing_phase_EA2.py for EA2 (NEAT)

Analysis

The following scripts are available for analysis:

• analyse_training_phase.py: analyzes fitness and diversity, creating a 2x2 grid of line plots for mean fitnesses and distribution of best fitness in each run for both EAs & both enemy sets. It also performs statistical tests on the data.
• analyse_testing_phase.py: takes data from the folder gain_res_EA1 or gain_res_EA2, creates a boxplot, and performs statistical tests on the gain data.

Controller scripts

Running solutions on Evoman

• controller_task2.py

Constructing Controllers for Evoman framework

• demo_controller.py & neat_controller.py: these scripts can be used to construct a controller for the evoman framework based on EA1 & NEAT output.

Folder structure

• EA1_files: contains results & all files for EA1 (Baseline EA).
• EA2_files: contains results & all files for EA2 (NEAT).
• figures: contains all figures used in the Task 2 Report