For a Reinforcement Learning class, I worked on a few algorithms :
- Policy Iteration
- Value Iteration
- SARSA
- Q-Learning
to work on the OpenAI gym Cliff Walking problem (for SARSA and Q-Learning) and Sutton's Reinforcement Learning book Grid World exercice (for Policy Iteration and Value Iteration).
python main.py {RD, VI, PI, SARSA, QL}
With {RD: Random, VI: Value Iteration, PI: Policy Iteration, SARSA: SARSA, QL: Q-Learning}
Use python main.py -h
to know more.
For Policy Iteration and Value Iteration, plots will appear, showing a map for each move (LEFT, RIGHT, UP, DOWN) colored when the given move is the best for the square. For SARSA and Q-Learning, plots will appear, showing the final reward after each episode. The parameters have been tuned so that the learning works (reward increase along the episodes).