• Python v3.6.5
• anaconda

1. Run conda create --name a4
2. Run source activate a4
3. Run conda install --yes --file requirements.txt
4. Run pip install gym pyglet

In this experiment, convergence and performance of value iteration and policy iteration are compared for 3 different MDPs, including:

1. FrozenLake-v0
2. FrozenLake8x8-v0
3. Taxi-v2

Reproduce the results by running python analysis/mdp.py

A Q-learner reinforcement learning algorithm was applied to the "Toy Text" environments. You can reproduce the results by running:

• python frozen_lake/q_learning.py
• python taxi/q_learning.py