Reinforcement Learning in Reversi game with Weight Agnostic Neural Network.
Based on article weightagnostic.github.io and my other repository Reversi-RL.
python wann_train.py -p p/reversi_5_4.json -n 8
python wann_test.py -p p/reversi_5_4.json -r 1000 -i champions/reversi_5_4.out -v True
Fitness may be interpreted as accuracy
[***] Fitness: [0.48 0.45 0.46 0.8 0.77 0.75]
[***] Weight Values: [-2. -1. -0.5 0.5 1. 2. ]
[***] Fitness: [0.58 0.56 0.55 0.4 0.43 0.43]
[***] Weight Values: [-2. -1. -0.5 0.5 1. 2. ]
[***] Fitness: [0.51 0.49 0.49 0.55 0.54 0.54]
[***] Weight Values: [-2. -1. -0.5 0.5 1. 2. ]