Cleaners

The cleaners combine several filters into higher-level cleaning strategies.

CNCNOSCleaner

Bases: BaseEstimator

CNC-NOS cleaner using the filters available in this project.

Parameters:

Name	Type	Description	Default
base_filters	sequence or None	Optional sequence of estimators or (name, estimator) pairs used as the base ensemble. When omitted, a heterogeneous default ensemble is built from the filters shipped with this repository.	None
base_neighbors	int	Number of neighbors used by the base neighbor-driven filters.	3
score_neighbors	int	Number of neighbors used to compute the weighted noise score.	5
cv	int	Number of folds used by the default cross-validated base filters.	10
metric	str	Distance metric used by the neighbor-based components.	"minkowski"
p	int	Minkowski power parameter, only used when metric="minkowski".	2
max_iter	int	Maximum number of cleaning iterations.	10
stagnation_patience	int	Number of consecutive low-improvement iterations tolerated before stopping.	2
wns_tol	float	Minimum improvement in mean weighted noise score to reset the stagnation counter.	1e-4
final_filtering	bool	Reserved flag for an optional final conservative pass. The current implementation keeps this pass disabled.	False
final_filtering_min_fraction	float	Minimum candidate fraction required to activate the optional final pass.	0.2
min_class_count	int	Minimum number of samples per class that must remain after cleaning.	2
random_state	int	Seed used to initialise the stochastic components.	33
n_jobs	int or None	Parallelism forwarded to the distance-based base filters.	None

Notes

The cleaner combines a candidate selection stage, a weighted noise score and a consensus/majority relabelling step before optionally removing the remaining candidates.

fit(X, y)

Fit the cleaner and cache the cleaning history and masks.

fit_resample(X, y)

Fit the cleaner and return the cleaned subset.

get_filter_report()

Return a dictionary with the main cleaning diagnostics.

Summary of a CNC-NOS cleaning run.

Per-iteration diagnostics for CNC-NOS.

What to look at in the report

• keep_mask: samples preserved after cleaning.
• relabel_mask: samples whose label was changed.
• remove_mask: samples removed from the final cleaned set.
• history: per-iteration trace of the cleaning process.