This is a machine learning project that applies a general-purpose Actor-Critic Reinforcement Learner (RL) to instances of a puzzle type known as Peg Solitaire. For a complete description of the game, see Wikipedia. A RL system consists of an agent and an environment, where the agent houses all of the core RL processes, while the environment contains everything else. The figure below provides a high-level view of the system that is implemented in this project. This system consists of the RL system (i.e. agent) composed of an actor and critic, and the SimWorld, which incorporates the environment and all knowledge about states and their relationships in that environment. As shown in the diagram, it also houses a structure representing the actual player of the game.
The main file initiates the agent and the environment and consists of pivot parameter values used to solve specied problems in the project description. The figures below show an example of the learning plot for one of the learning processes and the visualization of the final episode of this process.
| Visualization of last episode | Learning plot |
|---|---|
 |
 |