Questions about mcts4j code

[chwflhs]

Hi, I have read your code of mcts4j (Monte-Carlo_tree_search) carefully and found it quite useful for me, it is really a nice work.

However, I tested the program and found I cant get the correct answer. I think there still some problem in the code.

To do the test, I printed the execute steps of tic-tac-toe. The steps are given in the attached figure. Results varies in different tests, but many have similar problem:

In the attached figure (test 1), the problem is in turn 7.the correct move for O should be (1;2), but O moved (0;1) instead. So X moved (1;2) and win.

In the attached figure (test 2), the problem is in turn 6.the correct move for O should be (2;2), but ) moved (0;0) instead. Then in turn 7, the correct move for X should be (2;2), but X moved (1;0).

From the test result, I think the problem is related with the selection function. Currently i'm not sure of the exactly problem location and don't konw how to fix it so i write to ask your help.

[chwflhs]

I have found 1 possible reason for this, but maybe not all the reason, given in attached figure.

The setTerminal step: I understand it is used to visit all the tree without reiteration, however, next time when doTransition function execute, all the nodes is already flagged as terminal? need to be reset?

[avianey]

thanks for the feedback, I'll look into this ;-)

[avianey]

rewrote the code... but havn't look into this while rewriting... As of today, setTerminal only means its a leaf Node corresponding to end of game But terminal should also be optimized to hold the info that the whole subtree has been explored/expanded (to avoid reiteration)