Feature-selector orchestration, MNPO-facing config, method contracts, and stable result surfaces.

Package source: tabnetics.feature_selection

Package overview

Feature selection sub-package (Phase 2+3+6).

Stable exports

  • class FeatureSelector (class) - Source. Advanced Feature Selector with two strategies: 1. mnpo_portfolio (default): MNPO (Nash Multi-Portfolio Optimization) selection. 2. legacy_voting: legacy weighted ensemble voting.
  • class FeatureSelectionResult (class) - Source. Comprehensive result object for feature selection process.
  • class FeatureSelectorConfig (class) - Source. Top-level configuration for FeatureSelector.
  • class OracleConfig (class) - Source. MNPO oracle controls (Phase 6A, T-R-180).
  • METHOD_REGISTRY (constant) - Source. Module-level constant exported by the package surface.
  • class MethodSpec (class) - Source. Specification for a single feature selection method.
  • def get_method_weights() -> dict[str, float] (function) - Source. Return {key: legacy_weight} for all registered methods.
  • def get_experimental_keys() -> set[str] (function) - Source. Return the set of keys whose maturity is ‘experimental’.
  • class MethodContract(ABC) (class) - Source. Execution contract for one feature-selection method.
  • class FeatureSelectorMethodContract(MethodContract) (class) - Source. Thin adapter that routes the contract to an existing selector method.
  • def build_default_method_contracts(selector) -> Dict[str, MethodContract] (function) - Source. Build contracts for all registry-backed methods with callable handlers.

Module details

tabnetics.feature_selection.__init__

Source file

Feature selection sub-package (Phase 2+3+6).

No top-level public symbols are exported directly from this module.

tabnetics.feature_selection.base

Source file

  • class FeatureSelector (class) - Source. Advanced Feature Selector with two strategies: 1. mnpo_portfolio (default): MNPO (Nash Multi-Portfolio Optimization) selection. 2. legacy_voting: legacy weighted ensemble voting.

tabnetics.feature_selection.config

Source file

Configuration dataclasses for FeatureSelector.

  • DEFAULT_SELECTOR_PENALTY_MAP (constant) - Source. Module-level constant exported by the package surface.
  • class OracleConfig (class) - Source. MNPO oracle controls (Phase 6A, T-R-180).
  • class MNPOConfig (class) - Source. Mirror-descent Nash portfolio optimisation parameters.
  • class StabilityConfig (class) - Source. Stability selection, cluster-stability, decorrelated-stability, and IPSS parameters.
  • class WrapperConfig (class) - Source. Wrapper refinement and iterative-pruning parameters.
  • class MulticlassConfig (class) - Source. One-vs-All, ECOC, NSC, and class-Pareto multiclass parameters.
  • class CopulaConfig (class) - Source. D-vine copula knockoff (DTDCKe) parameters.
  • class PrefilterConfig (class) - Source. Prefilter blend configuration for feature pool reduction.
  • class ScreeningConfig (class) - Source. Tier 2 interaction-aware screening configuration.
  • class EvaluationConfig (class) - Source. Multi-classifier evaluation proxy configuration.
  • class MethodConfig (class) - Source. Per-method hyper-parameters for mRMR, k-TSP, and HSIC Lasso.
  • class FeatureSelectorConfig (class) - Source. Top-level configuration for FeatureSelector.

tabnetics.feature_selection.result

Source file

FeatureSelectionResult dataclass — comprehensive output of the feature selection pipeline.

  • class FeatureSelectionResult (class) - Source. Comprehensive result object for feature selection process.

tabnetics.feature_selection.registry

Source file

Canonical registry of all feature selection methods.

  • class MethodSpec (class) - Source. Specification for a single feature selection method.
  • METHOD_REGISTRY (constant) - Source. Module-level constant exported by the package surface.
  • def get_method_weights() -> dict[str, float] (function) - Source. Return {key: legacy_weight} for all registered methods.
  • def get_experimental_keys() -> set[str] (function) - Source. Return the set of keys whose maturity is ‘experimental’.

tabnetics.feature_selection.contracts

Source file

Method execution contracts for feature-selection methods.

  • class MethodContract(ABC) (class) - Source. Execution contract for one feature-selection method.
  • class FeatureSelectorMethodContract(MethodContract) (class) - Source. Thin adapter that routes the contract to an existing selector method.
  • def build_default_method_contracts(selector) -> Dict[str, MethodContract] (function) - Source. Build contracts for all registry-backed methods with callable handlers.

Documentation and webpages on this site are generated from authoritative internal sources using a combination of deterministic rules and generative AI. Errors are possible. Please report issues via GitHub Discussions or email [email protected].