This is the class project repo for ECE 239AS (Reinforcement Learning) Spring 2020 taught by Lin Yang.
This project is an re-implementation of the Progressive Neural Networks proposed in 2016 by Google DeepMind in PyTorch.
Team members are: Gaohong Liu, Jintao Jiang, Hengda Shi
This code is tested under python 3.7.7, dependencies can be installed with:
pip install -r requirements.txtsrc/common/params.py is the parameter setting file that documents all hyper-parameters used in the code.
Run the program with python src/main.py to begin the training process.
The default configuration utilizes 16 processes to accelerate the training process. No GPU is currently configured in the code. You could possibly move the PyTorch Tensors to GPU for training, but since the environments are rendered on CPU, this might not result in a better performance due to the CPU-GPU communication overhead.