This project is a simplified implementation of the AlphaZero algorithm using Monte Carlo Tree Search (MCTS) and a neural network. It demonstrates how self-play reinforcement learning can be used to train an agent from scratch without human data.
- AlphaZero-style training loop (self-play → training → iteration)
- Monte Carlo Tree Search (MCTS) implementation
- Neural network for policy + value prediction
- Supports multiple games:
- TicTacToe
- Connect Four
- Modular and extensible design (easy to plug in new games)
- The agent plays games against itself (self-play)
- MCTS is used to explore possible moves
- The neural network learns:
- Policy (best move probabilities)
- Value (winning chances)
- The model improves over multiple iterations
- Python
- NumPy
