Intelligent Human-Computer interface.

This repository is detailed extensively in the accompanied Research Paper.

How to Navigate this Repository

CustomDQN

The current version of the Deep Q-Learning Network implimentation. It provides a convience wrapper for training an agent in any enviornment.

gym_environment_tests

DQN algorithm applied to games from the popular benchmark Gym from OpenAI.

saved_scores

Visualizations of various model performance on different tasks.

future_models

RL models still in production that have no guarantee of their effectiveness.

code_references

Code samples found elsewhere on github, they may be used as a reference when updating files in future_models

How to get started

Install Python

sudo apt-get install python3.6

OpenAI's Universe environments are only supported on Linux and Mac distributions (does not work on Windows). If you only have a Windows OS you have the option to run CustomDQN on the OpenAI's Gym environments (see gym_environment_tests). I recommend using Python3.6 to execute the code in this repository.

Install Module Dependencies

Once you have python installed, you will need to install the required modules via pip:

pip install matplotlib numpy tensorflow keras gym universe

In order to use gym_enviornment_tests/LunarLander you will need to pip install gym[box2d] box2d-py

Clone Repository Tree using Git

git clone https://github.com/nathanShepherd/Intelligent-Interface.git

Execute Agent in Linux Terminal

sudo python mwob_Agent.py

Or you can observe the Agent control a Lunar Lander

python gym_environment_tests/LunarLander/lunarLander-CustomDQN.py

TODO:

• Use DenseNet to improve classification accuracy

• Use a CRF or RNN/LSTM to help estimate the Q-Function relative to the current point in time

• Augment memory for efficient and prioritized experience replay