Design for a generalized framework with options for double dqn, duel dqn, and prioritized experience replays.
- run.py - main facilitation of training and testing (hyperparameters are set here)
- options.py - object for storage of parameters (parameters and model requirements vary between environments)
- memory.py - structures for storing experiences in a memory replay (both standard queue and prioritized experience replay)
- Agents - both are nearly identical - separated for more efficient testing and cleaner individual codes
- agent.py - agent that contains the network, and generates actions based on states; leverages discrete states (ie linear states)
- conv_agent.py - agent that supports convolutional networks (ie image states)
- Models - both are available for significant modification based on network - separated for testing convenience
- model.py - for linear models
- conv_model.py for convolutional models