Labs and Homeworks for Reinforcement Learning course, MSc AI @ UvA 2018/2019.
Solutions and implementation from Davide Belli and Gabriele Cesa.
Total Grade: 102.75 %
Lab 1: Tabular Solutions Methods (grade 24.5/26)
Material covered in the lectures and in Sutton and Barto, chapters 2-7
Topics:
- Policy Evaluation
- Policy Iteration
- Value Iteration
- Monte Carlo Prediction
- Monte Carlo Control
- Temporal Difference learning
- Q-learning
- Sarsa
Environments:
- GridWorld
- Blackjack
- WindyGridWorld
Lab 2: Approximate Solutions Methods (grade: 31/25)
Material covered in the lectures and in Sutton and Barto, chapters 9-13
Topics:
- Deep Q-Network
- Experience Replay
- Semi-Gradient vs True Gradient
- Policy Gradient
- Policy Network
- MC Reinforce
- Actor-Critic
- Deep Reinforcement Learning + REINFORCE
- Deep Reinforcement Learning + Actor-Critic
Environments:
- CartPole
- CartPoleRaw (raw image observations)
Homework 1 (grade: 19.5/20)
Topics:
- Introduction to RL
- Exploration
- MDP
- Dynamic Programming
- Monte Carlo
- Temporal Difference learning (theory)
- Temporal Difference learning (application)
- Maximization Bias
- Model-based RL
- Contraction Mapping
- Banach's Fixed Point Theorem
Homework 2 (grade: 18.5/20)
Topics:
- Gradient Descent Methods
- Basis Functions
- Geometry of linear value-function approximation
- Neural Networks in RL
- REINFORCE
- Compatible Function Approximation Theorem
- Natural Gradient
Copyright
Copyright © 2019 Davide Belli.
This project is distributed under the MIT license. Please follow the UvA regulations governing Fraud and Plagiarism in case you are a student.