synthcity.plugins.core.models.mbi.graphical_model module
- class GraphicalModel(domain, cliques, total=1.0, elimination_order=None)
Bases:
object- belief_propagation(potentials, logZ=False)
Compute the marginals of the graphical model with given parameters
Note this is an efficient, numerically stable implementation of belief propagation
- Parameters
potentials – the (log-space) parameters of the graphical model
logZ – flag to return logZ instead of marginals
- Return marginals
the marginals of the graphical model
- calculate_many_marginals(projections)
- Calculates marginals for all the projections in the list using
Algorithm for answering many out-of-clique queries (section 10.3 in Koller and Friedman)
This method may be faster than calling project many times
- Parameters
projections – a list of projections, where each projection is a subset of attributes (represented as a list or tuple)
- Returns
a list of marginals, where each marginal is represented as a Factor
- datavector(flatten=True)
Materialize the explicit representation of the distribution as a data vector.
- fit(data)
- krondot(matrices)
- Compute the answer to the set of queries Q1 x Q2 X … x Qd, where
Qi is a query matrix on the ith attribute and “x” is the Kronecker product
This may be more efficient than computing a supporting marginal then multiplying that by Q. In particular, if each Qi has only a few rows.
- Parameters
matrices – a list of matrices for each attribute in the domain
- Returns
the vector of query answers
- static load(path)
- mle(marginals)
Compute the model parameters from the given marginals
- Parameters
marginals – target marginals of the distribution
- Param
the potentials of the graphical model with the given marginals
- project(attrs)
- Project the distribution onto a subset of attributes.
I.e., compute the marginal of the distribution
- Parameters
attrs – a subset of attributes in the domain, represented as a list or tuple
- Returns
a Factor object representing the marginal distribution
- static save(model, path)
- synthetic_data(rows=None, method='round')
Generate synthetic tabular data from the distribution. Valid options for method are ‘round’ and ‘sample’.
- greedy_order(domain, cliques, elim)
- variable_elimination(factors, elim)
run variable elimination on a list of (non-logspace) factors
- variable_elimination_logspace(potentials, elim, total)
run variable elimination on a list of logspace factors