In this repository, I have collected the differents reinforcement learning algorithm developed during Udacity Machine Learning Engineer Nanodegree Program.
-
DP_Methods.ipynb: Some implementations of many classical dynamic programming algorithms to solve the FrozenLake environment. -
DeepQ-Learning_Algorithm.ipynb: Teaching an agent to play games using Deep$Q$ -Learning Algorithm. To ilustrate, I have used this algorithms to teach an agent to play Cart-Pole. -
MC_Methods.ipynb: Teaching an agent to play Blackjack using differents Monte Carlo Methods. -
TD_Methods.ipynb: Implementation of many Temporal-Difference (TD) methods for solving Markov Decision Process (MDPs). For ilustrate, and testing the differents algorithms I have solved the CliffWalking environment.
In DDPG_ActorCritic_Methodfolder, there are an Actor-Critic Methods implementations, an similar scheme that I have used to teach a quadcopter to take off (see RL_Quadcopter
repository).
The enviroments used to apply the differents reinforcement learning algorithm are described in the differents notebook, similarly the algorithms used to solve them.
These environments have been implemented through Gym. It is a toolkit for developing and comparing reinforcement learning algorithms.
The instructions of using of each algorithm is described in its notebook.