smx.graph.builder

PredicateGraphBuilder: construct a directed predicate graph.

Edges are added between consecutive predicates ordered by a ranking metric inside each bag, and weights are accumulated across bags. Bidirectional edges are resolved by keeping only the higher-weight direction.

Classes

PredicateGraphBuilder

Build a directed predicate graph from bags and metric rankings.

Module Contents

class smx.graph.builder.PredicateGraphBuilder(random_state: int = 42, show_details: bool = True, var_exp: bool = False, pca_info_dict: Dict | None = None)

Build a directed predicate graph from bags and metric rankings.

Edge weights derive from the ranking metric of the source predicate in each bag. When the same directed edge appears in multiple bags, the weights are accumulated (summed). Bidirectional edges (A→B and B→A) are resolved by:

• keeping the edge with the higher accumulated weight;

• breaking ties randomly.

Parameters

random_stateint, default 42

Seed for random tie-breaking of bidirectional edges.

show_detailsbool, default True

Print details about identified and removed bidirectional edges.

var_expbool, default False

When True, multiply edge weights by the PC1 explained variance of the source predicate’s spectral zone. Requires pca_info_dict.

pca_info_dictdict, optional

{zone_name: {'variance_explained': float, ...}} as returned by smx.zones.aggregation.ZoneAggregator (pca_info_ attribute). Required when var_exp=True.

random_state = 42

show_details = True

var_exp = False

pca_info_dict = None

build(bags_result: Dict[str, Dict[str, pandas.DataFrame]], predicate_ranking_dict: Dict[str, pandas.DataFrame], metric_column: str = 'Covariance') → networkx.DiGraph

Build and return the directed predicate graph.

Parameters

bags_resultdict

Bags as returned by smx.predicates.bagging.PredicateBagger.

predicate_ranking_dictdict

{bag_name: DataFrame(['Predicate', metric_column])} as returned by a smx.predicates.metrics.BasePredicateMetric subclass.

metric_columnstr, default 'Covariance'

Name of the metric column in predicate_ranking_dict.

Returns

nx.DiGraph

Directed graph with 'weight' edge attributes.