Evolves an agent through tournament self-play using NEAT-Python on SlimeVolley task.
- set NEAT parameters in
config-feedforward
- set training parameters in
train.py
- run
python3 train.py
See output logs in log/neat-slimevolley
for final model .gif, saved checkpoints, and network graphs.
Uses feature engineering, and hit reward annealing to speed up convergence.