_Tensorflow (version 0.12.1)(https://www.tensorflow.org/install/)
_ALE (https://github.com/mgbellemare/Arcade-Learning-Environment)
This has been done using Python 2.7. Using Python 3.5+ will require to use Tensorflow 1.0+ and might require to change some lines of the script (some Tensorflow functions have been renamed or deprecated on the newer versions).
The script to run to train the network is DeepQLearninUP.py. It uses the rom breakout.bin (can be set in the python script). Results will be saved on logs_atari.