Here an overview of what happened before....
* CNN monster with billions of parameters wo search relies on ~billions of RL games
* parallel BestFirstMiniMax-Search on cpu with ann evaluation on gpu
* parallel BestFirstMiniMax-Search on gpu with ann evaluation on gpu
* cpu based alpha beta search with gpu ann eval
* fork of Zeta v099 but with neural networks
* fork of Zeta v098 but with neural networks