This library is a high speed implementation of Monte Carlo tree search guided by a policy and critic (based on transformer or Xgboost) for exploration. The tree is represented with a vectorize hash table (with pandas). Rollouts are performed in batch after a warm-up which discovers the first tree nodes.
This code can run around 100k rollouts in 10 seconds which is suitable for online drafting in MOBA games