This PR handles refactoring code to be more testable / cleaner.
Prior, all features exist but are hard to work with.
Edit: 10/14/2019
Redoing the entire MDP data structures. Now that we have tried testing continuous and discrete models on many different types of environments, we now have a better perspective on what these data structures need to do. Also we know know that memory management is going to be a core issue, and will make this directly into the structure.
Added:
Action object: For validating action sizes / dims, bundling important info
State: For validating state sizes / dims, bundling important info
Bounds: For determining the dtypes of its parent object, and determining if the object is discrete
Edit: 10/15/2019
Added:
MDPDataset: Cleaner and uses new data structures.
MDPStep: Better naming, cleaner also uses new data structures.
Edit: 10/21/2019
fast_rl is now unstably compatible with the fastai fit functions. This makes me super happy.
Edit: 10/23/2019
DDPG and DQN's seem to be working. Next step is to allow this to pass some tests and we will pull
This PR handles refactoring code to be more testable / cleaner. Prior, all features exist but are hard to work with.
Edit: 10/14/2019 Redoing the entire MDP data structures. Now that we have tried testing continuous and discrete models on many different types of environments, we now have a better perspective on what these data structures need to do. Also we know know that memory management is going to be a core issue, and will make this directly into the structure.
Added:
Edit: 10/15/2019
Added:
Edit: 10/21/2019
fast_rl is now unstably compatible with the fastai fit functions. This makes me super happy.
Edit: 10/23/2019
DDPG and DQN's seem to be working. Next step is to allow this to pass some tests and we will pull