DjAzDeck / SPG

Sample Policy Gradient

Sample Policy Gradient (SPG)

A complete implementation of deterministic policy gradient algorithms

Introduction

This repository contains four determinstic policy gradient algorithm implementations and serves as the backbone of my MSc Thesis research project with title:

Exploring Deep Reinforcement Learning for Continuous Action Control

A great amount of work has been put in order to deliver an easy-to-understand and comprehensive implementation of the algorithms. The interested reader should take a look at the complete text (link tba) in order to understand fully the scope of this research.

Setup

For an easy use, we recomment to create a new python3 virtual environment and activate it.

python3 -m venv SPG python=3.8

Install the requirements

pip3 install -r requirements.txt

Train the Agents

All you have to do is choose the environments and the parameters you wish to test on the main.py file and execute the script passing the name argmument -n. Example of usage:

python main.py -n 1mil --cuda

Environments

The PyBullet environments are already installed from the requirements but for MuJoCo environments, you need to follow the official Installation Guide.

Evaluation

In order to evaluate the final agents, you can run the evaluate_spg.py script. Example of usage:

python evaluate_spg.py -m <weights_path> -r True

With the above line you can also record the agent performing on the environments. It requires a virtual graphical display to be installed beforehand.

sudo apt-get install xvfb