monte carlo tree search question

[elcabesa]

I started reading something about MCTS and I'm asking myself how it's possible to implement it.

I'm used to negamax recursive implementation where you don't explicitly create the tree in memory but you traverse the tree without allocating it.
Is this possibile with montecarlo? or a montecarlo simply create a bit tree in memory and you have to store it?

the fist examples I have found create a tree in ram.

thank you.

[AlvaroBegue]

Yes, in Monte Carlo Tree Search you need to store the tree in memory. Just to be clear, the random playouts that go from the leaves to the end of the game don't need to be kept in memory.

[elcabesa]

yes, I understood this.

I have seen that an attempt of creating a monte carlo chess engine has been done in a thesis called MCC, derived from stockfish. it's still possibile to find it? Unofrunately I wasn't able to found the sources.

[Ajedrecista]

Hello Marco:

yes, I understood this.

I have seen that an attempt of creating a monte carlo chess engine has been done in a thesis called MCC, derived from stockfish. it's still possibile to find it? Unofrunately I wasn't able to found the sources.

I guess that you know about this paper of MCC:

Monte Carlo Chess

This paper contains brief pieces of code.

Regards from Spain.

Ajedrecista.

[elcabesa]

Yes I know about it. I want to know whether the full source code is available. I like to read code.

[Daniel Shawul]

yes, I understood this.

Yes I know about it. I want to know whether the full source code is available. I like to read code.

Why the hell would you want code when you can think/understand the algorithm for yourself? Do you think you can add some quick ELOs to your little chess engine with MCTS? The only reason you are even considering this now is probably because something other than alpha-beta have become a hot-topic. Well the answer is NO ... especially if your attitude is just for the Stockfish's version of MCTS. Are you interested in an 2200+ chess engine that plays chess and other variants with MCTS ? It is available for download here but no source code. https://sites.google.com/site/dshawul/home

[velmarin]

Ivanhoe has that option, and the code.

[elcabesa]

yes, I understood this.

Yes I know about it. I want to know whether the full source code is available. I like to read code.

Why the hell would you want code when you can think/understand the algorithm for yourself? Do you think you can add some quick ELOs to your little chess engine with MCTS? The only reason you are even considering this now is probably because something other than alpha-beta have become a hot-topic. Well the answer is NO ... especially if your attitude is just for the Stockfish's version of MCTS. Are you interested in an 2200+ chess engine that plays chess and other variants with MCTS ? It is available for download here but no source code. https://sites.google.com/site/dshawul/home

Thank you for your polite and gentle reply.
I'd like to view the code since I already had the paper and because I think it's easier to fully underStand an algorithm if I can read a code too.
Thank you very much, it's a pleasure to speak with a so gentle person

[Roger Brown]

yes, I understood this.

Yes I know about it. I want to know whether the full source code is available. I like to read code.

Why the hell would you want code when you can think/understand the algorithm for yourself? Do you think you can add some quick ELOs to your little chess engine with MCTS? The only reason you are even considering this now is probably because something other than alpha-beta have become a hot-topic. Well the answer is NO ... especially if your attitude is just for the Stockfish's version of MCTS. Are you interested in an 2200+ chess engine that plays chess and other variants with MCTS ? It is available for download here but no source code. https://sites.google.com/site/dshawul/home

Thank you for your polite and gentle reply.
I'd like to view the code since I already had the paper and because I think it's easier to fully underStand an algorithm if I can read a code too.
Thank you very much, it's a pleasure to speak with a so gentle person

Hello Marco,

Pure class.

Please continue in like vein.

Later.

[smatovic]

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