bnsl package

bnsl consists of several core classes, that is:

bnsl
    -DAG
    -Score
    -Estimator

DAG class

class bnsl.graph.DAG(incoming_graph_data=None, **kwargs)

Bases: networkx.classes.digraph.DiGraph

The graph class used in this project, inherited class nx.DiGraph. When it is initialized, it has to be acyclic.

property adj_df

The adjacent matrix in the form of pd.Dataframe of DAG. :param **kwargs:

Returns

pd.Dataframe

property adj_np

The adjacent matrix in the form of np.array of DAG.

Returns

np.array

do_operation(operation)
property genome

The genome of DAG

Returns

np.array

legal_operations()

Iterator, yield all legal operations like (‘+’, (u, v)), (‘-’, (u, v)), (‘flip’, (u, v)).

Yields

operation (‘+’, (u, v))

read(path: str, mode='edge_list', source='source node', target='target node')

Read DAG from csv file.There are two modes: edge_list, adjacent_matrix.

In the edge_list mode, the csv file looks like:

source node,target node
asia,tub
tub,either
either,dysp
either,xray

In the adjacent_matrix mode, the csv file looks like:

,asia,tub,either,smoke,lung,bronc,dysp,xray
asia,0,1,0,0,0,0,0,0
tub,0,0,1,0,0,0,0,0
either,0,0,0,0,0,0,1,1
smoke,0,0,0,0,1,1,0,0
lung,0,0,1,0,0,0,0,0
bronc,0,0,0,0,0,0,1,0
dysp,0,0,0,0,0,0,0,0
xray,0,0,0,0,0,0,0,0
Parameters

  • path – file path.

  • mode – edge_list mode or adjacent_matrix mode.

  • source – the source node column.

  • target – the target node column.

save(path: str, mode='edge_list')

Save the DAG into csv file.

Parameters

  • path – file path.

  • mode – edge_list mode or adjacent_matrix mode.

score(score_method, detail=False)

Calculate the score of the DAG.

Parameters

  • score_method – score criteria instance

  • detail – return a dictionary containing every local score. Default is False.

Returns

score of the DAG (or score, dict)

show()

Draw the figure of DAG

summary()

Print a summary to describe current dag.

Score class

class bnsl.score.MDL_score(data: pandas.core.frame.DataFrame)

Bases: bnsl.base.Score

MDL score.

reference:

Lam, W., & Bacchus, F. (1994). Learning Bayesian belief networks: An approach based on the MDL principle. Computational intelligence, 10(3), 269-293. Yuan, C., Malone, B., & Wu, X. (2011, June). Learning optimal Bayesian networks using A* search. In Twenty-Second International Joint Conference on Artificial Intelligence.

likelihood(Nijk: numpy.ndarray)

Calculate the likelihood.

Nijk is a table-like array.
    q1  q2  q3
r1  10  12  30
r2  50  60  17
(qi are the states of parents)
(ri are the states of x)
Parameters

Nijk – state table

Returns

likelihood.

local_score(x: str, parents: tuple)

Calculate local score.

References

Kitson, N.K., Constantinou, A.C., Guo, Z., Liu, Y. and Chobtham, K., 2021. A survey of Bayesian Network structure learning. arXiv preprint arXiv:2109.11415.

Parameters

  • x – str: Node.

  • parents – tuple: Parents of node.

Returns

Local score

Return type

float

class bnsl.score.BIC_score(data: pandas.core.frame.DataFrame, **kwargs)

Bases: bnsl.base.Score

BIC score class, BIC score is minus MDL score.

local_score(x, parents)

Calculate local score of BIC score.

Parameters

  • x – str: node

  • parents – tuple: parents of node.

Returns

BIC score

class bnsl.score.BDeu_score(data: pandas.core.frame.DataFrame, equivalent_sample_size=10, **kwargs)

Bases: bnsl.score.MDL_score

BDeu score.

local_score(x, parents)

Calculate local score.

Parameters

  • x – node.

  • parents – parents of node.

Returns

BDeu score.