This is a Deep Q Network running tensorflow.
It is aimed at running open ai gym.

Network Description:
2 relu hidden layers with 100 neurons each (overkill)
Adam optimizer with .01 learning rate
.95 Discount rate for Q Learning
a random chance of taking an exploratory action
chance decays over time

Note:
I have not yet optimized those parameters

Code availible here:
https://github.com/nburn42/nburn42_deep_q

reproducing should be as easy as running: python deep_q.py (assuming gym, box2d, tensorflow, and numpy are installed and you are using python2)

Nathan Burnham nburn42@gmail.com