sc_a3c: another attempt for SC2 AI

This is the A3C method applied by Deepmind in 2017 to train SC2 AI, and A3Chas been proved successful for minigames.

dqfd_new.py: old DQfD version algorithm, proved unsuccessful

a3c_tl: A3C on continious controlling based on tensorlayer. Copied from TensorLayer's Github.

a3c_discrete: A3C for discrete decision problems. Copied from Morvan Zhou.

a3c_sc: My implementation of A3c on pysc2.

Up till 1.1, most work has been finished. Check todo.txt to find out improvements.

En Taro Tassadar, and enjoy yourself.

References:

A3C totorials and codes: https://morvanzhou.github.io/tutorials/machine-learning/reinforcement-learning/6-1-A-AC/