The "TS" in "MCTS" stands for "Tree Search". If you limit yourself to Monte Carlo simulations but you only store the results of the first ply, I wouldn't call that MCTS.smatovic wrote:a simple mcts implementation does not need to store the whole tree in memory, just the first ply with win/loss counters....afaik in GO mcts they use the uct extension, which stores expanded nodes in memory...http://chessprogramming.wikispaces.com/UCT
monte carlo tree search question
Moderator: Ras
-
AlvaroBegue
- Posts: 932
- Joined: Tue Mar 09, 2010 3:46 pm
- Location: New York
- Full name: Álvaro Begué (RuyDos)
Re: monte carlo tree search question
-
smatovic
- Posts: 3536
- Joined: Wed Mar 10, 2010 10:18 pm
- Location: Hamburg, Germany
- Full name: Srdja Matovic
Re: monte carlo tree search question
my fault, it would be then only Monte Carlo simulations/playouts...
-
elcabesa
- Posts: 858
- Joined: Sun May 23, 2010 1:32 pm
Re: monte carlo tree search question
I just found some free time to start experimenting with MCTS after a few monts from my last post.
I'm just implementing the autoplayer to play the simulation stage of the mcts algorithm.
I'm just letting the engine play a game against himself at a fixed depth from the simulation initial position, is this right? the simulation in MC should be random playout, but if a do a real random game I think it would never give me good result. do you think that a search at a fixed depth, for exampe 10 is the right way?
I'm just implementing the autoplayer to play the simulation stage of the mcts algorithm.
I'm just letting the engine play a game against himself at a fixed depth from the simulation initial position, is this right? the simulation in MC should be random playout, but if a do a real random game I think it would never give me good result. do you think that a search at a fixed depth, for exampe 10 is the right way?