jackbenn / oware Goto Github PK

Untrained model classes should be specified in a separate file (models.py?). This would contain a base class and a number of subclasses, mostly different neural networks but also random models that don't learn. When a new model is instantiated, it should be based on the class.

Trained models instances (called something else?) should saved in a directory, as they are.

Agent learning should also have options; these can be changed from one game to the next. These would include different values of epsilon but also different algorithms (e.g., MC, TD(n)).