Estimators¶

class lgbn.estimators.GreedyEquivalentSearch(score=None, max_iter=1000000, eps=1e-09)¶

Greedy Equivalent Search structure learning algorithm.

The Greedy Equivalent algorithm learns the structure of a Bayesian network that maximizes the given score. The search procedure starts with an empty graph. Edges are added until no more increase the score and then removed until no further operation increases the score. Equality of operations and thus networks is defined by equivalence classes. An equivalence class contains all networks which have the same edges regardless of orientation. This algorithm is reasonably fast when used with a decomposable score which can be cached.

See 3 for a detailed description of the Greedy Equivalent Search algorithm.

Note

This implementation requires a decomposable score, although there exist other implementations that work with non-decomposable scores.

References

3: D. M. Chickering, “Optimal Structure Identiﬁcation With Greedy Search,” Journal of Machine Learning Research, vol. 3, no. Nov 2002, p. 48, Nov. 2002.

get_params(deep=True)¶

Get parameters for this estimator.

Parameters: deep (bool, default=True) – If True, will return the parameters for this estimator and contained subobjects that are estimators.
Returns: params – Parameter names mapped to their values.
Return type: dict

search()¶: Search the space of posible models for the one that maximizes the score of this estimator.

set_params(**kwargs)¶

Set the parameters of this estimator.

The method works on simple estimators as well as on nested objects (such as Pipeline). The latter have parameters of the form <component>__<parameter> so that it’s possible to update each component of a nested object.

Parameters: **params (dict) – Estimator parameters.
Returns: self – Estimator instance.
Return type: estimator instance

class lgbn.estimators.GreedyHillClimbing(score=None, start_net=None, max_iter=1000000, eps=1e-09, random_state=None)¶

Greedy Hill Climbing structure search algorithm.

The Greedy Hill Climbing algorithm learns the structure of a Bayesian network that maximizes the given score. The search procedure starts with an initial network, which defaults to a fully disconnected network. Edges are added, removed or have their direction reversed one at a time until no more modifications increase the overall score of the network. This algorithm is reasonably fast when used with a decomposable score which can be cached.

See p. 40 in 2 for a detailed description of the Greedy Hill Climbing algorithm. The source refers to Greedy Hill Climbing as Max-Min Hill Climbing.

Note

This implementation requires a decomposable score, although there exist other implementations that work with non-decomposable scores.

References

2: I. Tsamardinos, L. E. Brown, and C. F. Aliferis, “The max-min hill-climbing Bayesian network structure learning algorithm,” Mach Learn, vol. 65, no. 1, pp. 31–78, Oct. 2006, doi: 10.1007/s10994-006-6889-7.

get_params(deep=True)¶

Get parameters for this estimator.

Parameters: deep (bool, default=True) – If True, will return the parameters for this estimator and contained subobjects that are estimators.
Returns: params – Parameter names mapped to their values.
Return type: dict

search()¶: Search the space of posible models for the one that maximizes the score of this estimator.

set_params(**kwargs)¶

Set the parameters of this estimator.

The method works on simple estimators as well as on nested objects (such as Pipeline). The latter have parameters of the form <component>__<parameter> so that it’s possible to update each component of a nested object.

Parameters: **params (dict) – Estimator parameters.
Returns: self – Estimator instance.
Return type: estimator instance

class lgbn.estimators.K2Search(score=None, ordering=None, eps=1e-09)¶

K2 structure learning algorithm.

The K2 algorithm learns the structure of a Bayesian network that maximizes the given score. The search procedure is guided by a given topological ordering of the network. In that ordering, if node x comes before node y, then node y can never be a parent of node x. This vastly reduces the search space resulting in a significant speedup, even without using caching.

See 1 for a detailed description of the K2 algorithm.

Note

This implementation requires a decomposable score, although there exist other implementations that work with non-decomposable scores.

References

1: G. F. Cooper and E. Herskovits, “A Bayesian method for the induction of probabilistic networks from data,” Mach Learn, vol. 9, no. 4, pp. 309–347, Oct. 1992, doi: 10.1007/BF00994110.

get_params(deep=True)¶

Get parameters for this estimator.

Parameters: deep (bool, default=True) – If True, will return the parameters for this estimator and contained subobjects that are estimators.
Returns: params – Parameter names mapped to their values.
Return type: dict

search()¶: Search the space of posible models for the one that maximizes the score of this estimator.

set_params(**kwargs)¶

Set the parameters of this estimator.

The method works on simple estimators as well as on nested objects (such as Pipeline). The latter have parameters of the form <component>__<parameter> so that it’s possible to update each component of a nested object.

Parameters: **params (dict) – Estimator parameters.
Returns: self – Estimator instance.
Return type: estimator instance

class lgbn.estimators.ScoreSearchEstimator(score=None, eps=1e-09)¶

A structure estimator using score-based search.

Note

This class is just a general interface, it cannot actually be used.

Estimators¶

lgbn

Navigation

Related Topics