This work is part of the Udacity Deep Reinforcement Learning Nanodegree third assignment, which consists on solving the Tennis environment.
To setup your local Python environment for running Unity environments checkout the instructions on this Github repository. On this work we'll use PyTorch to build the networks. On requirements.txt you'll also find some other packages required.
This work does not require to install Unity, the environment is already been built, and you can download it from the link below:
Then you must place the environment inside the env folder, or update the path on the notebook, if you wish to reproduce the report.ipynb.
You should follow report.ipynb for the detailed implementation process. The models folder holds all the model files that was used, and utils folder has the support files, such as the noise and replay buffer implementations. On the agent.py file the main agent is implemented, the one responsible for creating and training both the actor and critic networks.
You can use the trained model by loading the parameters from model_parameters folder to both the actor and critic network, and then acting on the environment.
This work can be improved by testing another noise functions, as well as implementing Prioritized Experience Replay.