This project is my second project on Udacity's Deep Reinforcement Learning Nanodegree (Facebook PyTorch Nanodegree Scholarship Phase 3). This project uses a Deep Deterministic Policy Gradients (DDPG) Network to train a double-jointed arm to move to target locations.
The environment the first verison of the Reacher enviroment, which is an extension of Unity’s UnityEnvironment. The Reacher environment has different versions for Linux, Windows (32-bit), Windows (64-bit) and MacOSX. The following screenshot shows the environment set-up:
The state is a numpy array of length 33.
The action is how the agent reacts to the environment.
The environment is considered ‘solved’ when the moving average (arithmetic mean) score over the last 100 episodes is greater than or equal to 30.0.
Complete the following steps interact with this project:
- Determine your operating system (OS)
- Determine your number of bits (only if you are using Windows) - click here for help from Microsoft
- Click the relevant hyperlink from the following list
- Place the file you downloaded in this repo's folder on your computer and unzip (or decompress) the file click here for help
- Open
Continuous_Control.ipynbusing Jupyter Notebook (or Jupyter Lab) - Click Cell>Run All
- Be patient as the magic happens