Post by steadyeddie on Dec 18, 2016 6:13:58 GMT -8
At the most basic level my Nodes just form a tree which is the T in MCTS.
It escaped my notice for some time that in some games (but by no means all) a position can be reached from more than other node. I've got no idea how important it is, but I figured what I needed to do was cache Node from a hash of the position, and save memory. What you end up with after you do this, is not a Tree, but a Directed Acyclic Graph. So the T is, in fact, bogus.
I don't have edges, but i know people do. Instead I'm just storing references to the children, alongside the list of moves.
It took me a while to realise I needed to store the score for both sides, as opposed to just my score, so I could simulate the other player playing the move that is best for them. That's needed for multi-player games, and for non zero-sum games.
It escaped my notice for some time that in some games (but by no means all) a position can be reached from more than other node. I've got no idea how important it is, but I figured what I needed to do was cache Node from a hash of the position, and save memory. What you end up with after you do this, is not a Tree, but a Directed Acyclic Graph. So the T is, in fact, bogus.
I don't have edges, but i know people do. Instead I'm just storing references to the children, alongside the list of moves.
It took me a while to realise I needed to store the score for both sides, as opposed to just my score, so I could simulate the other player playing the move that is best for them. That's needed for multi-player games, and for non zero-sum games.
