| Project | Implemented components | Status |
|---|---|---|
1_TicTacToe |
Environment, random/human/minimax agents, tournament, policy-gradient and actor-critic agent packages | Source-backed |
2_Connect4 |
Environment/vectorized environment, human/search agents, tournament, DQN/Double DQN/dueling DQN packages, trainer and test entrypoint | Source-backed |
3_Othello |
Directory only | Scaffold |
Chess |
Directory only | Scaffold |
Connect4 Q-networks consume a multi-channel board representation and use residual blocks. Its agent implementations use replay buffers and TensorFlow gradient updates; traditional agents provide non-learning baselines. The projects do not ship trained checkpoints or match-result claims.