Comparative study of Monte Carlo, SARSA & Q-Learning on custom MiniGrid environments. Phase-conditioned state representation, one-shot event-based reward shaping, Optuna hyperparameter search. TD methods: 100% success; MC: 61% on the hard env. RL course HW2, Reichman University.
python machine-learning reinforcement-learning deep-learning monte-carlo jupyter-notebook q-learning hyperparameter-optimization sarsa minigrid bayesian-optimization gymnasium reward-shaping temporal-difference optuna tabular-rl minigrid-environment
-
Updated
Jun 22, 2026 - Jupyter Notebook