Hacker News new | past | comments | ask | show | jobs | submit login

the reasoning for that thinking is simply because MCTS is comparable to DFS while something like minimax is comparable to BFS.

Of course, because MCTS searches in a depth-first manner, the deeper the depth the less effective it is, because you have less time for your rollouts (as your rollouts are more expensive).

But not many board games (if we are limiting ourselves to this example) will have thousands of moves until end of game -- they may, however, have an exponentially large breadth of moves.

For a game like Go, MCTS could simulate thousands of rollouts in the amount of time it would take minimax to evaluate two plies ahead.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: