Atari game - deep Q learning neural network (Keras based)
The model used here is mostly following the paper by DeepMind: Playing Atari with Deep Reinforcement Learning
Demo gif after training.
max_iter = 10,000,000
steps between evaluation = 100,000
learning rate = 0.00025
steps between target network copying = 10,000
experience replay memory size = 100,000
gamma = 0.99 (discount factor)
optimizer = RMSProp
