facebook-api.txt

// a representation of probability that supports either symbolic or concrete
// values
struct probability;

// a conditional probability table
struct cpt;
struct bayesian_network;
struct compiled_bayesian_network;

// specifies how many values each variable in the bayesian network can take
// shape[0] gives the # of values for var 0
typedef unsigned int* shape;

// create a Bayesian network data structure
bayesian_network* make_bayesian_network(unsigned int num_vars,
            shape s,
            cpt * cpts,              // list of conditional probability tables
            );

// create a CPT
// probabilities[i] gives the conditional probability at index i
//
// For example: if vars = [0, 1] and shape = [3,2], meaning there are two variables in 
// this CPT (vars 0 and 1) and var 0 has 3 values and var 1 has 2 values, then
// i=0 is state (var 0 = 0, var 1 = 0)
// i=1 is state (var 0 = 0, var 1 = 1)
// i=2 is state (var 0 = 1, var 1 = 0)
// ...
cpt* make_cpt(shape s, unsigned int* vars, unsigned int num_vars, probability* probabilities)

// a struct that maps symbol probabilities to concrete probabilities
struct symbol_table; 

// generate an empty symbol table
symbol_table* make_symbol_table();

// sets a probability in the symbol table 
void set_probability(symbol_table* st, unsigned int idx, double probability);

// make a symbol probability that will be substituted later during inference
// the ID uniquely identifies this probability
probability* make_symbolic_prob(unsigned int id);

// make a concrete probability
probability* make_concrete_prob(double val);


// compile bayesian network for amortized inference
compiled_bayesian_network* compile_bayesian_network(unsigned int num_vars, cpt * cpts, unsigned int num_cpt);

// Computes the joint marginal for the subset of variables `query_vars`
double** marginal(compiled_bayesian_network* bn, symbol_table* st, unsigned_int* query_vars, unsigned_int num_vars);