Current location - Education and Training Encyclopedia - University rankings - What algorithm is Monte Carlo tree?
What algorithm is Monte Carlo tree?
Monte Carlo Tree Search (MCTS) will gradually build an asymmetric tree. It can be divided into four steps and iterated repeatedly:

(1) Select

Starting from the root node, that is, the decision-making situation R, choose a node T that needs to be expanded most urgently; Case R is the first node to check. If the checked node has a move M that has not been evaluated, then the new situation obtained after the checked node executes M is that we need to expand T; If all feasible moves in the checking situation have been evaluated, a feasible move with the largest ucb value is obtained by using ucb formula, and the new situation generated by this move is checked again; If the checked situation is that the game has ended, directly execute step 4; Through repeated checking, we finally get the last checked situation C and the unevaluated movement M at the bottom of the tree, and execute step 2.

(2) expansion

Add a child node to Scenario C that exists in memory at this time. This child node is obtained by scenario c executing move M, which is t.

(3) Simulation

Starting from situation T, the two sides began to settle at random. Finally, a result (win/lose) is obtained to update the winning rate of T node.

(4) Back propagation

T After the simulation, its parent node C and all its ancestor nodes update the winning rate in turn. The winning rate of a node is the average winning rate of all its children. It starts from T and propagates back to the root node R, so the winning percentage of all nodes on the path will be updated.