Filtering API

TL;DR

filter_model compares an observed sparse EdgeTable with a fitted independent null model and returns sparse edge subsets with p-values and expectations.

Entry point

from menobis.filtering import filter_model

result = filter_model(
    edges,
    family=ModelFamily.ME,
    constraint=Constraint.STRENGTH,
    fit=fit,
    alpha=0.05,
    tail="two-sided",
)

Strength-cost filtering also needs coordinates:

result = filter_model(
    edges,
    family=ModelFamily.ME,
    constraint=Constraint.STRENGTH_COST,
    fit=fit,
    coord_x=x,
    coord_y=y,
)

Options

Option	Default	Meaning
alpha	0.05	significance level
tail	two-sided	upper, lower, or two-sided
correction	none	none, bonferroni, or fdr
detect_absent	False	scan zero-weight candidate pairs
min_occupation	0.5	absent-pair occupation threshold
min_expected	0.0	absent-pair expected-weight threshold
max_absent	None	cap absent output

Result shape

result.upper
result.lower
result.compatible
result.absent_lower

Each field is a FilteredEdges object with:

Field	Meaning
edges	sparse edge table
upper_pvalue	upper-tail p-values
lower_pvalue	lower-tail p-values
expected	null expected occupation
occupation	null probability of positive occupation

Supported constraints

Filtering is for independent grand-canonical nulls: strength, strength-cost, strength-edges, strength-degree, degree-events, and sparse custom/partial Poisson rates through lower-level wrappers.