Batch Monte Carlo Tree Search

Making inferences with a deep neural network on batch of states is much faster GPU than making one state after another. We build this property to propose Monte Carlo Tree Search algorithms using batched inferences. Instead either search tree or transposition table we use both in the same algorithm. The contains results while statistics Search. also analyze multiple heuristics that improve search: $$\mu $$ FPU, Virtual Mean, Last Iteration and Second Move heuristics. They are evaluated for game Go MobileNet network.