reinforcement-learning-game

This project contains a self trained ai game from open GymAI library using reinforecment learning that will be presented as a Machine Learning workshop for beginners at the Greek National ECE Student Conference (SFHMMY) of 2019.

Installation

You need to install all dependencies first using

pip install -r requirements.txt

Install the GymAI games with,

pip install gym[atari]

or with

pip install git+https://github.com/Kojoley/atari-py.git

Presentations

• A how to Python presentation in Python-workshop.pdf is made for beginners to learn the python syntax and how to use basic python libraries, such as numpy and pandas. If this is your first time with python, start from here.
• The complete workshop presentation is in BuildYourFirstAIGame.pdf.

File list

• In mountain_car.py a Q-learning algorithm is used to train an agent to play and win the mountain car game, from gym, openAI library, as shown here.
• In space_invaders.py the target is the same. An agent has to learn to win the Atari Space Invaders game. The game environment can be found here. In this case, the environment is much more complex and a simple Q-learning algorithm can't solve quickly and easy enough. Deep learning is used in combination with a Q-table, written in Keras.
• check_env.py is a simple script to check the existence and the real view of the space invaders game. It initializes the game and plays randomly forever, until the process is killed.