Extension stage in MCTS

Discussion of chess software programming and technical issues.

Moderators: hgm, Harvey Williamson, bob

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Post Reply
sovaz1997
Posts: 166
Joined: Sun Nov 13, 2016 9:37 am

Extension stage in MCTS

Post by sovaz1997 » Sat Dec 01, 2018 10:17 am

How is the extension stage in MCTS implemented? Do we expand by adding a random move and do simulations or do we add all the moves and do simulations for all the moves?

Thanks!
Zevra chess engine
Binary, source and description here: https://gitlab.com/sovaz1997/zevra2/tags

Gerd Isenberg
Posts: 2110
Joined: Wed Mar 08, 2006 7:47 pm
Location: Hattingen, Germany

Re: Extension stage in MCTS

Post by Gerd Isenberg » Mon Dec 03, 2018 8:05 pm

Guess you mean expansion strategy ... using the UCB1 formula, see
https://www.youtube.com/watch?v=UXW2yZndl7U

trulses
Posts: 34
Joined: Wed Dec 06, 2017 4:34 pm

Re: Extension stage in MCTS

Post by trulses » Thu Dec 06, 2018 8:34 pm

In "your typical MCTS" you use your tree policy to pick moves while you're in already expanded nodes. Like Gerd says this might be UCB1 or something to that effect. When you eventually reach a position that's outside your tree you then add it as a leaf node and run your evaluation function and back that value up the tree. This evaluation function might be a neural net like in A0 or random rollouts to terminal states what have you. No need to add any more moves in this leaf node.

sovaz1997
Posts: 166
Joined: Sun Nov 13, 2016 9:37 am

Re: Extension stage in MCTS

Post by sovaz1997 » Thu Dec 06, 2018 10:58 pm

Thank! I seem to understand how this works. :)
Zevra chess engine
Binary, source and description here: https://gitlab.com/sovaz1997/zevra2/tags

Post Reply