Libmcts

Monte Carlo Simulation library

Requirements

• UnitTest++ for tests (expects headers and lib to be in /usr/{lib,include})

Build

$ git clone --recursive https://github.com/pandaant/libemcts

# compile lib
$ cd libmcts
$ make all

# build documentation (requires doxygen)
$ make doc

# if you want to run tests
$ cd test 
$ make all run

A compiled version of the doxygen documentation can be found here

Usage

The usage will be demonstrated using an example implementation for the game of Rock, Paper, Scissors (found in the test folder).

using namespace mcts;

// define the nodetype
typedef typename INode<RockPaperScissors, RPSConfig>::node_t node_t;

// define the selection strategy
typedef typename ISelectionStrategy<RockPaperScissors, RPSConfig>::sstrategy_t sstrategy_t;

// define the mcts with the types for your implementation
typedef typename MCTS<RockPaperScissors, RPSConfig, RPSNode<RockPaperScissors,RPSConfig>, RPSResult<RockPaperScissors,RPSConfig>> mcts_t;

// create needed variables
RockPaperScissors context;
RPSConfig *conf;

Player p1, p2;
ActionType::Enum p1a, p2a;

mcts_t *mcts;
IBackpropagationStrategy *bp_strat;
sstrategy_t *select_strat;
sstrategy_t *move_select_strat;

// creates an initial game state to start simulations from
int game = 1;
p1 = Player("mark");
p2 = Player("simon");
++p1.won; ++p2.lost;
p1a = ActionType::PAPER;
p2a = ActionType::ROCK;

context = RockPaperScissors(p1, p2, p1a, p2a, game);

// initialize strategies
bp_strat = new AvgBackpropagationStrategy();
select_strat = new MaxValueSelector<RockPaperScissors,RPSConfig>();
move_select_strat = new MaxValueSelector<RockPaperScissors,RPSConfig>();

conf = new RPSConfig(bp_strat, select_strat, move_select_strat, false);
mcts = new mcts_t(context, conf);

// iterate 1000 times
for (int i = 0; i < 1000; ++i)
    mcts->iterate();

// get the result of the simulations
RPSResult<RockPaperScissors, RPSConfig> result = mcts->result();

// free variables
delete mcts;
delete conf;
delete move_select_strat;
delete bp_strat;
delete select_strat;

When creating your own abstactions

The derived context object has to implement the following functions to interact properly with the library. The transition function returns a vector of new contexts that are derived from the current context.

class Context{
	vector<Context> transition();  
}

The used config object has to implement the following functions to interact properly with the library. The nb_gen returns the rng generator used for internal use.

The generator itself needs to implement the () operator to generate a new number.

class Config{
	generator\_t *nb_gen();
}