Implementation of:
-DQN -> http://home.uchicago.edu/~arij/journalclub/papers/2015_Mnih_et_al.pdf
-Double DQN -> http://arxiv.org/abs/1509.06461
-Dueling network -> http://arxiv.org/abs/1511.06581
-Prioritized experience replay -> http://arxiv.org/pdf/1511.05952v3.pdf
-Deep Attention Recurrent Q-Network -> https://arxiv.org/pdf/1512.01693v1.pdf (Coming next)
-Asynchronous Methods for Deep Reinforcement Learning -> https://arxiv.org/pdf/1602.01783v2.pdf (Coming next)
Currently, the config file parameters are tunned in order to learn to play to flappy bird (https://github.com/sourabhv/FlapPyBird).
Dependencies:
-Tensorflow
-python 2.7
-openCV
-pyQtGraph
-numpy
-pqdict
11/20/2016 - learning DDQN & Dual network OK
11/23/2016 - PER integrated but learning is much slower