npretor_udacity_deepRL

This Deep Q network maps the 37 state spaces of the Unity environment to a best policy for choosing the 4 possible actions in the environment. More detail is given in the report. The majority of the project code is taken from this Unity project: https://github.com/udacity/Value-based-methods

1. Installation

1. Clone the repo

   git clone https://github.com/npretor/npretor_udacity_deepRL
   cd npretor_udacity_deepRL
   

2. Install conda if not installed

3. a. Install dependancies: Installing the unity dependencies was a pain (running OS X Big Sur). That can be another blog post. Install non-unity reqs using:

   conda create --name deepqn --file package-list.txt
   

   If using jupyterlab:

   python3 -m ipykernel install --user --name=deepqn
   source activate deepqn
   python3 -m jupyterlab 
   

   If using bash

   source activate deepqn
   

3. Training -

This runs without the Weights and Biases module. Use the jupyter notebook for weights and biases tracking, but you will need to login and config wandb yourself.

python3 Navigation_Training.py

4. Demo

python3 Navigation_Demo.py

Report

• Report