Reinforcement Learning

This repository puts together different reinforcement learning projects done under a university course

MazeMDP

Modeling of a Maze problem as a Markov Decision Process and solution using classical dynamic programming and value iteration.

MountainCar

Shows how to train an agent in the Gym environment Mountain cart using a SARSA algorithm trained using stochastic gradient descent with Nesterov acceleration and using as function approximation for the value function a linear regressor with Fourier basis.

A detailed analysis of the results can be found under the docs folder.

Deep Architectures

Here it is possible to find an implementation similar to RLLib for training and testing Reinforcement Learning methods in a versatile way. It shows how to implement DDPG and DQN under this framework to solve tasks in a Gym environment.