mrftools.BruteForce module

BruteForce class for exact inference of marginals and maximizing states.

class mrftools.BruteForce.BruteForce(markov_net)

Bases: object

Object that can do inference via ugly brute force. Recommended only for sanity checking and debugging using tiny examples.

compute_z()

Compute the partition function by explicitly summing energy of all possible states. This is extremely expensive for anything but tiny Markov nets.

Returns:the partition function (normalizing constant) of the distribution Return type:float

entropy()

Compute the entropy of the distribution by explicitly accounting for every possible state.

Returns:entropy Return type:float

map_inference()

Compute most probable state configurations, i.e., maximum a posteriori (MAP) inference by explicitly trying every possible state.

Returns:a matrix of one-hot indicator vectors for the maximizing states of all variables Return type:ndarray

pairwise_marginal(var_i, var_j)

Compute the joint marginal probabilities between two variables.

Parameters:

• var_i (object) – first variable to marginalize over
• var_j (object) – second variable to marginalize over

Returns:

matrix representing the marginal probabilities of the two variables

Return type:

ndarray

unary_marginal(var)

Compute the marginal probabilities of a variable.

Parameters:var (object) – variable whose marginals will be computed Returns:vector of marginal probabilities for var Return type:array