gfn.gym.helpers.bayesian_structure.priors

Classes

BasePrior	Base class for the prior over graphs p(G).
EdgePrior	Base class for the prior over graphs p(G).
ErdosRenyiPrior	Base class for the prior over graphs p(G).
FairPrior	Base class for the prior over graphs p(G).
UniformPrior	Base class for the prior over graphs p(G).

Module Contents

class gfn.gym.helpers.bayesian_structure.priors.BasePrior(num_variables=None)

Bases: abc.ABC

Base class for the prior over graphs p(G).

Any subclass of BasePrior must return the contribution of log p(G) for a given variable with num_parents parents. We assume that the prior is modular.

Parameters:

num_variables (int (optional)) – The number of variables in the graph. If not specified, this gets populated inside the scorer class.

__call__(num_parents)

_log_prior = None

_num_variables = None

abstract property log_prior

property num_variables

class gfn.gym.helpers.bayesian_structure.priors.EdgePrior(num_variables=None, beta=1.0)

Bases: BasePrior

Base class for the prior over graphs p(G).

Any subclass of BasePrior must return the contribution of log p(G) for a given variable with num_parents parents. We assume that the prior is modular.

Parameters:

num_variables (int (optional)) – The number of variables in the graph. If not specified, this gets populated inside the scorer class.

beta = 1.0

property log_prior

class gfn.gym.helpers.bayesian_structure.priors.ErdosRenyiPrior(num_variables=None, num_edges_per_node=1.0)

Bases: BasePrior

Base class for the prior over graphs p(G).

Any subclass of BasePrior must return the contribution of log p(G) for a given variable with num_parents parents. We assume that the prior is modular.

Parameters:

num_variables (int (optional)) – The number of variables in the graph. If not specified, this gets populated inside the scorer class.

property log_prior

num_edges_per_node = 1.0

class gfn.gym.helpers.bayesian_structure.priors.FairPrior(num_variables=None)

Bases: BasePrior

Base class for the prior over graphs p(G).

Any subclass of BasePrior must return the contribution of log p(G) for a given variable with num_parents parents. We assume that the prior is modular.

Parameters:

num_variables (int (optional)) – The number of variables in the graph. If not specified, this gets populated inside the scorer class.

property log_prior

class gfn.gym.helpers.bayesian_structure.priors.UniformPrior(num_variables=None)

Bases: BasePrior

Base class for the prior over graphs p(G).

Any subclass of BasePrior must return the contribution of log p(G) for a given variable with num_parents parents. We assume that the prior is modular.

Parameters:

num_variables (int (optional)) – The number of variables in the graph. If not specified, this gets populated inside the scorer class.

property log_prior