I have implemented some of famous algorithms for making skills using the Option framework in Reinforcement Learning
for running the code from the command line, you should select the environment and skill acquisition algorithm. For grid world domain the default world is 6-room, but you can change it according the domains defined in calss sharif.ce.isl.rl.graph.environment.Grids. The parameters for SCC algorithm is set in the constructor of class sharif.ce.isl.rl.graph.algorithm.SCC, but you can change it.
-
running SCC algorithm for making skills: -alg sharif.ce.isl.rl.graph.algorithm.SCC -env sharif.ce.isl.rl.graph.environment.GridWorldEnv -iniStat 1,1 -finStat 30,23 -numEpisode 100
-
running Q-learning algorithm: -alg sharif.ce.isl.rl.graph.algorithm.core.QLearningAgent -env sharif.ce.isl.rl.graph.environment.TaxiDriverEnv -numEpisode 100
For using this code please cite at least to one of the following papers: @article{kazemitabar2017graph, title={A graph-theoretic approach toward autonomous skill acquisition in reinforcement learning}, author={Kazemitabar, Seed Jalal and Taghizadeh, Nasrin and Beigy, Hamid}, journal={Evolving Systems}, pages={1--18}, year={2017}, publisher={Springer Berlin Heidelberg} }
@article{nasrin2013novel, title={A Novel Graphical Approach to Automatic Abstraction in Reinforcement Learning}, author={Nasrin Taghizadeh, Hamid Beigy}, journal={Robotics and Autonomous Systems}, volume={61}, pages={821--835}, year={2013} }
@inproceedings{ghafoorian2013automatic, title={Automatic Abstraction in Reinforcement Learning Using Ant System Algorithm.}, author={Ghafoorian, Mohsen and Taghizadeh, Nasrin and Beigy, Hamid}, booktitle={AAAI Spring Symposium: Lifelong Machine Learning}, year={2013} }
for any question, please contact with nasrin.taghizadeh@gmail.com.