select_relevant_features()

select_relevant_features() runs Boruta all-relevant feature selection on an existing feature matrix. Use this when you already have X (for example from extract_sequence_features()) and a target y.

Function Usage

select_relevant_features(
    X,
    y,
    *,
    problem_type="regression",
    n_iter=50,
    perc=100.0,
    boruta_alpha=0.01,
    boruta_two_step=False,
    random_state=42,
    verbose=False,
)

R / Literature Parameter Mapping

Sequenzo	R Boruta	Notes
n_iter	maxRuns	Outer Boruta iterations
perc	shadow percentile logic	Default 100.0
boruta_alpha	pValue	Default 0.01
boruta_two_step	mcAdj / two-step path	Sequenzo default False
problem_type	implicit from y	"regression" or "classification"

Underlying engine: BorutaPy + sklearn random forest (Gini importance), not R ranger permutation importance.

Entry Parameters

Parameter	Required	Type	Description
X	✓	ndarray / DataFrame	Feature matrix (n × p). Column names used if DataFrame.
y	✓	array-like	1D target vector of length n.
problem_type	✗	str	"regression" or "classification".
n_iter	✗	int	Boruta iterations (default 50).
perc	✗	float	Percentile of shadow importances (default 100.0).
boruta_alpha	✗	float	Significance level (default 0.01).
boruta_two_step	✗	bool	BorutaPy two-step correction (default False).
random_state	✗	int / None	Random seed for the forest (default 42).
verbose	✗	bool	Print Boruta progress.

What It Returns

A dict with:

Key	Description
selected_mask	Boolean mask over columns (confirmed).
selected_indices	Integer indices of confirmed features.
selected_feature_names	Names of confirmed features.
tentative_mask	Boolean mask for tentative features (support_weak_).
tentative_indices	Indices of tentative features.
tentative_feature_names	Names of tentative features.
boruta_ranking	Boruta ranking vector (ranking_ from BorutaPy).
hit_counts	Reserved for Boruta implementations that expose hit counts; currently None with BorutaPy.
shadow_hit_counts	Reserved for Boruta implementations that expose shadow hit counts; currently None with BorutaPy.

If X is not a DataFrame, features are named X1, X2, …

Example

from sequenzo import extract_sequence_features, select_relevant_features

features = extract_sequence_features(seqdata, timing_bin_width=12.0)
selection = select_relevant_features(
    features["X_full"],
    outcome,
    problem_type="regression",
    n_iter=50,
    boruta_alpha=0.01,
    boruta_two_step=False,
)

print(selection["selected_feature_names"])
print(selection["tentative_feature_names"])

With residualized outcome (manual)

Residualize y on controls before calling this function if you follow Bolano and Studer (2020); the pipeline does this automatically via residualize_target_with_controls.

R Counterpart

• Closest R function: Boruta::Boruta()
• Mapping note: Same algorithm family; different importance metric and RF backend. See Conceptual Guide.

Notes

• X must contain only finite values.
• For classification, invalid category codes (y < 0) raise an error.
• Requires PyPI package boruta (pip install sequenzo).
• Default random forest uses n_estimators=800 inside the Boruta wrapper unless you extend select_all_relevant_features_boruta with a custom estimator (not exposed on this entrypoint).
• With BorutaPy, hit_counts and shadow_hit_counts are None. Use boruta_ranking, selected_*, and tentative_* for diagnostics.

Authors

Code: Yuqi Liang

Documentation: Yuqi Liang

References

Bolano, D., & Studer, M. (2020). The link between previous life trajectories and a later life outcome: A feature selection approach.

Unterlerchner, L., Studer, M., & Gomensoro, A. (2023). Back to the features. Investigating the relationship between educational pathways and income using sequence analysis and feature extraction and selection approach. Swiss journal of sociology, 49(2), 417-446.