My Minor Thesis project investigating the viability of merging attributes of reinforcement learning and genetic algorithms to combat poor generalization in reward sparse enviroments.
Python Version 3.7
Import dependencies in scripts/requirements.txt
Test a model using scripts/environment/MarioEnvironment.py
Train a new model using scripts/ga/TrainMarioGA.py
When training, do not use odd population numbers and the minimum population must be larger than '5'
Due to the method used to select parents and ensure genetic performance isn't lost this restriction is in place
See scripts/ga/util/MarioGAUtil.py for details
Settings for Population are in scripts/ga/TrainMarioGA.py
Settings for Agents are in scripts/nn/setup/AgentParameters.py
The following projects were used to help me develop this codebase: