Task 2 — Reinforcement Learning (Chef’s Hat Gym)

Student + Variant

This aligns with the assignment brief requirement that ID mod 7 = 6 must incorporate a generative AI component and assess its impact.

Environment: ChefsHatGYM (local Room-based multi-agent game loop)
Opponents: 3 × RandomAgent
My agent: PGGenAIAgent (REINFORCE policy gradient, PyTorch)
Generative AI Augmentation:
- A tiny generative opponent model learns action-to-action transition probabilities from observed opponent moves
- The predicted opponent action distribution is appended to the agent’s state vector as an extra feature

python examples/train_task2_variant6.py