This is the code to solve gym-woodoku, a reinforcement learning environment based on the woodoku game.
After training, you can modify the weight_path in the monitoring.py file to record the agent's gameplay.
The parameters for training are inside the woodoku_ppo_v2*.py file and can be modified as desired.
The code referenced
- Brown :
woodoku_ppo_v2_action_mask_combo.py - Pink :
woodoku_ppo_v2_action_mask.py- Increased the reward of Pink by 10x.
- Red:
woodoku_ppo_v2_action_mask.py - Bottom blue: invalid_action_penalty
- Top blue: Loaded the last weight of the bottom blue and trained as invalid_action_penalty